Дисертації з теми "Category classification"

In Australia, dentition (eruption of permanent incisors) is used as a proxy for age to define sheep meat quality. Lamb is defined as having no permanent incisors, hogget as having at least one incisor and mutton is defined as having two or more incisors. Classification of the carcase is done at the abattoir prior to the removal of an animal’s head. Recently, an Australian Senate inquiry into meat marketing reported that there was concern that substitution of hogget and mutton for lamb may be occurring in the industry. At present, no objective method is available that can be used for classifying sheep category. The general aims of this thesis were to i) evaluate whether chemical analysis of branched chain fatty acid (BCFA) content could be used as an objective tool to determine sheep age, ii) understand the effect that some production factors had on BCFA concentrations in Australian sheep and iii) develop new approaches (whether chemical and/or statistical) for determining sheep category (age). BCFAs are implicated as the main contributors to “mutton flavour”, often associated with the cooked meat of older animals. BCFAs are reported to increase with age, which suggests that chemical analysis of these compounds could be used as an objective method. Concentrations of three BCFAs (4-methyloctanoic (MOA), 4-ethyloctanoic (EOA) and 4- methylnonanoic (MNA) acids) were measured in a survey of fat samples taken from 533 sheep carcases at abattoirs in New South Wales, Victoria and Western Australia. This thesis shows that, on its own, chemical analysis of the BCFAs is not sufficient to discriminate lamb from hogget and mutton as pre-slaughter nutrition is a significant factor in classifying sheep using this approach. Uncertainty at the BCFA concentration ranges found in Australian sheep was determined to be high making it difficult to discriminate between sheep carcases of different ages based on the BCFA level. Fast gas chromatography was evaluated as the basis for a high throughput chemical technique but was not sufficiently sensitive for BCFA measurements. Solid-phase microextraction (SPME) was also found to be suitable for sampling 3-methylindole and p-cresol, compounds responsible for diet-related “pastoral flavour” in sheep fat, but further work is needed to validate this approach for measurement of these compounds in sheep fat. Statistical classification algorithms, when applied to the chromatograms measured for the 533 carcasses, showed great promise for predicting sheep category. Specifically, the random forests algorithm, when applied to mean-centred data, gave 100% predictive accuracy when differentiating between lamb, hogget and mutton. This approach could be used for the development of an objective method for determining sheep age and category, suitable for use by the Australian sheep meat industry.

Skeletal shape representations, in spite of their structural instabilities, have proven themselves as effective representation schemes for recognition and classification of visual shapes. They capture part structure in a compact and natural way and provide insensitivity to visual transformations such as occlusion and articulation of parts. In this thesis, we explore the potential use of disconnected skeleton representation for shape recognition and shape classification. Specifically, we first investigate the importance of contextual information in recognition where we extend the previously proposed disconnected skeleton based shape matching methods in different ways by incorporating category knowledge into matching process. Unlike the view in syntactic matching of shapes, our interpretation differentiates the semantic roles of the shapes in comparison in a way that a query shape is being matched with a database shape whose category is known a priori. The presence of context, i.e. the knowledge about the category of the database shape, influences the similarity computations, and helps us to obtain better matching performance. Next, we build upon our category-influenced matching framework in which both shapes and shape categories are represented with depth-1 skeletal trees, and develop a similarity-based shape classification method where the category trees formed for each shape category provide a reference set for learning the relationships between categories. As our classification method takes into account both within-category and between-category information, we attain high classification performance. Moreover, using the suggested classification scheme in a retrieval task improves both the efficiency and accuracy of matching by eliminating unrelated comparisons.

Image representation is in the heart of many computer vision algorithms. Different computer vision tasks (e.g. classification, detection) require discriminative image representations to recognize visual categories. In a nutshell, the bag-of-visual-words image representation is the most successful approach for object and scene recognition. In this thesis, we mainly revolve around this model and search for discriminative image representations. In the first part, we present a novel approach to incorporate spatial information in the BoVW method. In this framework, we present a simple and efficient way to infuse spatial information by taking advantage of the orientation and length of the segments formed by pairs of similar descriptors. We introduce the notion of soft-similarity to compute intra and inter visual word spatial relationships. We show experimentally that, our method adds important discriminative information to the BoVW method and complementary to the state-of-the-art method. Next, we focus on color description in general. Differing from traditional approaches of invariant description to account for photometric changes, we propose discriminative color descriptor. We demonstrate that such a color description automatically learns a certain degree of photometric invariance. Experiments show that the proposed descriptor outperforms existing photometric invariants. Furthermore, we show that combined with shape descriptor, the proposed color descriptor obtain excellent results on four challenging data sets.Finally, we focus on the most accurate color representation i.e. multispectral reflectance which is an intrinsic property of a surface. Even with the modern era technological advancement, it is difficult to extract reflectance information without sophisticated instruments. To this end, we propose to use the display of the device as an illuminant while the camera captures images illuminated by the red, green and blue primaries of the display. Three illuminants and three response functions of the camera lead to nine response values which are used for reflectance estimation. Results show that the accuracy of the spectral reconstruction improves significantly over the spectral reconstruction based on a single illuminant. We conclude that, multispectral data acquisition is potentially possible with consumer hand-held devices such as tablets, mobiles, and laptops

The process of acquiring language requires children to learn grammatical categories and apply these categories to new words. Researchers have proposed various explanations of this process in the form of algorithms and computational modeling. Recently, adaptation and selection models have been tested and applied as a possible explanation to the process of acquiring grammatical categories. These studies have proven promising, however, the external validity of this approach has not been examined by grammatically coding samples outside the training corpus. The current thesis applies an adaptation and selection model, which pauses the evolution of dictionaries after every thousand cycles to allow the tagging of 30 outside samples, which are then checked for tagging accuracy. The accuracy across the five training corpora by the six thousandth cycle averaged 76.75%. Additional research is needed to explore the effects of altering the parameters in the model.

Cette thèse porte sur la théorie de la discrimination multi-classe à marge. Elle a pour cadre la théorie statistique de l’apprentissage de Vapnik et Chervonenkis. L’objectif est d’établir des bornes de généralisation possédant une dépendances explicite au nombre C de catégories, à la taille m de l’échantillon et au paramètre de marge gamma, lorsque la fonction de perte considérée est une fonction de perte à marge possédant la propriété d’être lipschitzienne. La borne de généralisation repose sur la performance empirique du classifieur ainsi que sur sa "capacité". Dans cette thèse, les mesures de capacité considérées sont les suivantes : la complexité de Rademacher, les nombres de recouvrement et la dimension fat-shattering. Nos principales contributions sont obtenues sous l’hypothèse que les classes de fonctions composantes calculées par le classifieur ont des dimensions fat-shattering polynomiales et que les fonctions composantes sont indépendantes. Dans le contexte du schéma de calcul introduit par Mendelson, qui repose sur les relations entre les mesures de capacité évoquées plus haut, nous étudions l’impact que la décomposition au niveau de l’une de ces mesures de capacité a sur les dépendances (de la borne de généralisation) à C, m et gamma. En particulier, nous démontrons que la dépendance à C peut être considérablement améliorée par rapport à l’état de l’art si la décomposition est reportée au niveau du nombre de recouvrement ou de la dimension fat-shattering. Ce changement peut affecter négativement le taux de convergence (dépendance à m), ce qui souligne le fait que l’optimisation par rapport aux trois paramètres fondamentaux se traduit par la recherche d’un compromis
This thesis deals with the theory of margin multi-category classification, and is based on the statistical learning theory founded by Vapnik and Chervonenkis. We are interested in deriving generalization bounds with explicit dependencies on the number C of categories, the sample size m and the margin parameter gamma, when the loss function considered is a Lipschitz continuous margin loss function. Generalization bounds rely on the empirical performance of the classifier as well as its "capacity". In this work, the following scale-sensitive capacity measures are considered: the Rademacher complexity, the covering numbers and the fat-shattering dimension. Our main contributions are obtained under the assumption that the classes of component functions implemented by a classifier have polynomially growing fat-shattering dimensions and that the component functions are independent. In the context of the pathway of Mendelson, which relates the Rademacher complexity to the covering numbers and the latter to the fat-shattering dimension, we study the impact that decomposing at the level of one of these capacity measures has on the dependencies on C, m and gamma. In particular, we demonstrate that the dependency on C can be substantially improved over the state of the art if the decomposition is postponed to the level of the metric entropy or the fat-shattering dimension. On the other hand, this impacts negatively the rate of convergence (dependency on m), an indication of the fact that optimizing the dependencies on the three basic parameters amounts to looking for a trade-off

Variable screening techniques are fast and crude techniques to scan high-dimensional data and conduct dimension reduction before a refined variable selection method is applied. Its marginal analysis feature makes the method computationally feasible for ultra-high dimensional problems. However, most existing screening methods for classification problems are designed only for binary classification problems. There is lack of a comprehensive study on variable screening for multi-class classification problems. This research aims to fill the gap by developing variable screening for multi-class problems, to meet the need of high dimensional classification. The work has useful applications in cancer study, medicine, engineering and biology. In this research, we propose and investigate new and effective screening methods for multi-class classification problems. We consider two types of screening methods. The first one conducts screening for multiple binary classification problems separately and then aggregates the selected variables. The second one conducts screening for multi-class classification problems directly. In particular, for each method we investigate important issues such as choices of classification algorithms, variable ranking, and model size determination. We implement various selection criteria and compare their performance. We conduct extensive simulation studies to evaluate and compare the proposed screening methods with existing ones, which show that the new methods are promising. Furthermore, we apply the proposed methods to four cancer studies. R code has been developed for each method.

In part one we show that for a compact, metric, locally path-connected topological space X, the shape group of X - as defined in Foundations of Shape Theory by Mardesic and Segal - is isomorphic to the inverse limit of discrete homotopy groups introduced by Conrad Plaut and Valera Berestovskii. We begin by providing the reader preliminary definitions of the fundamental group of a topological space, inverse systems and inverse limits, the Shape Category, discrete homotopy groups, and culminate by providing an isomorphism of the shape and deck groups for peano continua. In part two we develop work and provide further classification of Sharkovskii topological groups, which we call Sharkovskii Groups. We culminate in proving the fact that a locally compact Sharkovskii group must either be the real numbers if it is not compact, or a torsion-free solenoid if it is compact.

This dissertation deals with the evolutionary optimization of ART neural network architectures. ART (adaptive resonance theory) was introduced by a Grossberg in 1976. In the last 20 years (1987-2007) a number of ART neural network architectures were introduced into the literature (Fuzzy ARTMAP (1992), Gaussian ARTMAP (1996 and 1997) and Ellipsoidal ARTMAP (2001)). In this dissertation, we focus on the evolutionary optimization of ART neural network architectures with the intent of optimizing the size and the generalization performance of the ART neural network. A number of researchers have focused on the evolutionary optimization of neural networks, but no research has been performed on the evolutionary optimization of ART neural networks, prior to 2006, when Daraiseh has used evolutionary techniques for the optimization of ART structures. This dissertation extends in many ways and expands in different directions the evolution of ART architectures, such as: (a) uses a multi-objective optimization of ART structures, thus providing to the user multiple solutions (ART networks) with varying degrees of merit, instead of a single solution (b) uses GA parameters that are adaptively determined throughout the ART evolution, (c) identifies a proper size of the validation set used to calculate the fitness function needed for ART's evolution, thus speeding up the evolutionary process, (d) produces experimental results that demonstrate the evolved ART's effectiveness (good accuracy and small size) and efficiency (speed) compared with other competitive ART structures, as well as other classifiers (CART (Classification and Regression Trees) and SVM (Support Vector Machines)). The overall methodology to evolve ART using a multi-objective approach, the chromosome representation of an ART neural network, the genetic operators used in ART's evolution, and the automatic adaptation of some of the GA parameters in ART's evolution could also be applied in the evolution of other exemplar based neural network classifiers such as the probabilistic neural network and the radial basis function neural network.
Ph.D.
School of Electrical Engineering and Computer Science
Engineering and Computer Science
Computer Engineering PhD

A análise dos sistemas de classificação do conhecimento na Filosofia e na Biblioteconomia sob um ponto de vista histórico-conceitual crítico com enfoque nos conceitos de classe, categoria e faceta foi desenvolvida a partir dos objetivos específicos de acompanhar e delinear a trajetória das classificações dos saberes (classes) e dos seres (categorias) à luz da Filosofia; averiguar as possíveis influências que as classificações filosóficas historicamente exerceram sobre as classificações bibliográficas tradicionais; e verificar de que modo os conceitos específicos de categoria, de classe e de faceta têm sido definidos no âmbito das classificações biblioteconômicas tradicionais. Trata-se de uma pesquisa do tipo exploratório-descritivo, embasada na literatura pertinente, oriunda de ambas as áreas do conhecimento em questão. A partir do levantamento bibliográfico construiu-se o referencial teórico, conduzindo à objetivação, análise, discussão e o aprofundamento do objeto. É possível afirmar que tanto os princípios das classificações filosóficas em relação aos saberes e em relação aos seres, quanto as próprias classificações que dividem os saberes e os seres, apresentam-se como construtos destinados a conhecer e disciplinar o conhecimento do ser e do saber. A influência das classificações filosóficas sobre as classificações bibliográficas reside no fato de que as classificações bibliográficas são adaptações das classificações do conhecimento ou das ciências. As bibliográficas, frequentemente, utilizam termos originários das filosóficas, ressignificando-os e transformando-os em ferramentas. Em síntese, pode-se afirmar que, a partir da Filosofia, absorvida em explicações teóricas, até a Biblioteconomia, preocupada com soluções instrumentais e aplicativas (pragmáticas), os esquemas de classificação bibliográfica propostos tem se beneficiado da aproximação dessas duas áreas de conhecimento. A revisão conceitual com respeito às noções de classe, categoria e faceta aponta, por parte dos estudiosos da Classificação, para a necessidade de um cuidado maior no desenho, no planejamento e na estruturação de sistemas de classificação, na modificação e especificação de tabelas de classificação e até de linguagens de indexação, com vista à sua adequação e relevância. As contribuições daí resultantes, delineadas numa perspectiva funcional-instrumental, poderão ajudar os profissionais a reconsiderar os seus sistemas de idéias e procedimentos em relação à construção e avaliação de linguagens de indexação e à classificação habitual de documentos. Caberá à Ciência da Informação avançar na fundamentação teórica do seu campo de aplicação, discutindo criticamente a sua base conceitual, atenta às práticas de uso em voga na Biblioteconomia e Documentação com relação a termos e conceitos. O estudo valida as hipóteses inicialmente levantadas, determinando que a consistência terminológica de termos como categoria, classe ou faceta contribui para aperfeiçoar a operacionalização do processamento do conhecimento em Ciência da Informação e que aos esquemas categoriais não cabe um valor neutro, uma vez que sempre favoreceram uma determinada concepção de mundo, em que pese a sua natureza objetivamente pragmática. Por fim, apresentam-se algumas perspectivas decorrentes da investigação, selecionando-se questões relevantes e de interesse para futuras pesquisas.
The analysis of knowledge classification systems in Philosophy and Librarianship by a critical historical-conceptual point of view with an approach in the concepts of class, category and facet was developed from specific objectives to accompany and outline the path of knowledge (classes) and beings (categories) classification in relation to Philosophy; to investigate the possible influences that the philosophical classification historically exerted on traditional bibliographical classifications; and to verify in which way the specific concept of category, class and facet have been defined in the ambit of traditional librarianship classifications. This research is of exploratory-descriptive type, based upon pertinent literature, resulting from both areas of knowledge in discussion. From the studied bibliographical material, a theoretical reference was built, leading to objectivation, analysis, discussion and a profound study of the object. Its possible to affirm that even the principles of philosophical classifications in relation to knowledge and beings, as well as the classifications that divide knowledge and beings, are presented as constructs, purposed to understand and to train the information of beings and knowledge. The influence of philosophical classifications on bibliographical classification is settled by bibliographical classifications, which are adaptations of knowledge or science. The bibliographical ones, frequently utilize terms originated from philosophical that are modified and transformed into tools. In synthesis, it may be stated that, from Philosophy, absorbed in theoretical explanations, even Librarianship, worried with instrumental and application (pragmatic) solutions, the proposed plans of bibliographical classifications have been beneficiated from the approximation of this two areas of knowledge. The conceptual review in relation to notions of class, category and facet needs a better attention in design, planning and in the structuring of classification systems, as pointed out by experts on Classification, in the modification and specification of classification tables and also indexation languages, aiming its adequacy and significance. The resulting contributions, outlined by a functional-instrumental perspective, would help professionals to reconsider their systems of ideas and procedures in relation to construction and evaluation of indexation languages and regular classification of documents. The information science will be responsible to develop theory fundamentation of its application field, discussing critically its conceptual basis, with attention to the practices in use by Librarianship and Documentation in relation to terms and concepts. The study validates the hypothesis initially described, determining that the terminological consistency of terms as category, class or facet contribute to improve the operacionalization of knowledge processing in Information Science and that to category schemes it doesnt fit a neuter value, since that always favored a certain conception of the world, in which predominates its pragmatic nature As result of the investigation, some perspectives are presented, selecting questions of extreme relevancy and interest for future researches.

In psychological research, many questionnaires use verbal response scales with vague linguistic terms (e.g., frequency expressions). The words’ meanings can be formalized and evaluated using fuzzy membership functions (MFs), which allow constructing distinct and equidistant response scales. The discriminatory power value of MFs indicates how distinct the functions and, hence, the verbal expressions are. The present manuscript interrogates the threshold of discriminatory power necessary to indicate a sufficient difference in meaning. Using an empirical validation procedure, participants (N = 133) estimated (1) three correspondence values for verbal expressions to determine MFs, and (2) similarities of words by pairwise comparison ratings. Results show a non-linear relationship between discriminatory power and similarity, and fuzzy MFs, as well as the searched-for threshold value for discriminatory power. Implications for the selection of verbal expressions and the construction of verbal categories in questionnaire response scales are discussed.

Um den Inhalt von Musik ausfindig zu machen, spielt ihre Gattungszugehörigkeit eine große Rolle. Dieses Gebiet ist in der Musikwissenschaft jedoch weniger erforscht als z. B. die Harmonie- oder Formenlehre. Von Interesse scheinen mir dabei insbesondere die Gattungen der Unterhaltungsmusik (Lieder, Tänze, Märsche) und verwandte Bereiche zu sein.

Textual data forms a popular form of communication; however, textual data is complex in nature as it is produced by humans. Given the huge amount of textual data currently available, it is essential to be able to mine this data automatically. Recent text mining efforts are making extensive use of knowledge bases, and this thesis pursues a similar effort. We however make use of Wikipedia to solve complex text mining tasks and current approaches do not make effective use of the category-article structure within Wikipedia. Particularly, we solve the problem of determining various topical threads in a document together with contextualization of social media content to disambiguate its various aspects. Experimental evaluations demonstrate the superiroty of our proposed methods when compared with state-of-the-art.

The assumption in economics of a representative agent is often made. However, it is a very rigid assumption. Hall and Jones (2004b) presented an economic model that essentially provided for a representative agent for each age group in determining the group's health level function. Our work seeks to extend their theoretical version of the model by allowing for two representative agents for each age—one for each of “Healthy” and “Sick” risk-factor groups—to allow for additional heterogeneity in the populace. The approach to include even more risk-factor groups is also briefly discussed. While our “extended” theoretical model is not applied directly to relevant data, several techniques that could be applicable were the relevant data to be obtained are demonstrated on other data sets. This includes examples of using linear classification, fitting baseline-category logit models, and running the genetic algorithm.

År 2019 var rekordår i transaktionsvolym på Stockholms kommersiella fastighetsmarknad där en mångfald av investerarkategorier var aktiva. De lägen där fastighetsinvesteringar görs kategoriseras kommunvis för fastighetstyper efter lägets attraktion på en skala som benämns AA, A, B, C, och D där AA är det mest attraktiva läget och D det minst attraktiva läget. Ett toppläge i en kommun kan inte jämföras med ett toppläge i en annan kommun trots att de båda kan ha samma klassificering, exempelvis A. Generellt för branschen är att de mest attraktiva lägena finns i Stockholms innerstad. Detta har väckt intresset för att finna samband mellan investerarkategorier och fastighetsinvesteringars lägen i Stockholms län som helhet samt mellan Stockholms kommun och Övriga kommuner i jämförelse.  Som en fördjupning utförs en analys i hur investeringspåverkande faktorer i ett A-läge kan skilja sig mellan olika kommuner, där A-lägen valts för att det finns mer data att tillgå än för AA-lägen. Metoden som använts för att finna samband mellan investerarkategorier och investeringars lägen i Stockholms län, Stockholms kommun och dess övriga kommuner har bestått av att kartlägga och analysera transaktionsdata över Stockholms län 2019. För att ta reda på vilka investeringspåverkande faktorer som finns för att investera i ett A-läge i olika kommuner i Stockholms län har en enkätundersökning som grundar sig på teori om investeringspåverkande faktorer genomförts.  Kartläggning av transaktionsdatan visar en spridd fördelning av investerarkategorierna i olika lägen både i Stockholms kommun och övriga kommuner. Exempel på vad kartläggningen visar är att de mest aktiva investerarkategorierna i AA-lägen var privata fastighetsbolag och övriga investerare samt att de opportunistiska fastighetsfonderna enbart genomförde transaktioner i B-lägen i Stockholms kommun. Enkätundersökningen visar att det finns vissa likheter i påverkande faktorer till att investera i industrifastigheterna mellan kommunerna. Faktorer som påverkade att investera i industrifastigheter i ett A-läge visar sig vara anknutna till samtliga nivåer av läget: mikroläge, närområde och närområdets attraktivitet. Vidare fanns en generell inställning att närhet till motorväg och hög teknisk standard var positiva faktorer för att investera i en industrifastighet i ett A-läge. Vidare framgår att påverkande faktorer för att investera i en fastighet för bostadsändamål har koppling till hur detaljplan i området möjliggör önskvärd användning, vilket på längre sikt kan innebära ett lyft för närområdet. Andra påverkande faktorer till att investera i en fastighet för bostadsändamål kan relateras till närområdet där samhällsservice och stadsmiljö ansågs attraktivt enligt investerarna. En intressant skillnad när det kommer till fastighetsinvesteringar för bostadsutvecklingsändamål är att de fysiska faktorerna i mindre grad verkade påverka investeraren i Stockholms kommun, jämfört med investeraren i Täby kommun där det ansågs vara väldigt viktigt.
2019 was a record year in transaction volume in Stockholm's commercial real estate market. A variety of investor categories were involved in transactions in Stockholm County. The locations where real estate investments are made are categorized by the municipality and its property type according to the location's attraction on a scale with AA, A, B, C, and D where AA is the most attractive location and D the least attractive location. An A-location in one municipality is not necessarily comparable with an A-location in another municipality, where the city in general has the most attractive locations. This has aroused interest in finding links between investor categories and the locations of real estate investments in Stockholm County as well as between Stockholm municipality and other municipalities in a comparison. As a deeper study, an analysis is carried out to explain how factors affecting real estate investment in an A-location can differ between different municipalities. Investments in A-locations are chosen to analyze since it no matter municipality is supposed to be an attractive location and A-locations have more available data than AA-locations. The method used to find links between investor categories and the locations of investments in Stockholm County, Stockholm Municipality, and its other municipalities has consisted of mapping and analyzing transaction data over Stockholm County 2019. To understand how factors affecting investments differ between A-locations in different municipalities in Stockholm County a survey based on the theory of investment influencing factors has been implemented.  The transaction data shows a variation of investor categories in different locations, both in Stockholm municipality and other municipalities. Examples of what the data shows are that the most active investor categories in AA-locations were private real estate companies and “other investors”, and that the opportunistic real estate funds only were involved in transactions in B-locations in Stockholm municipality. The survey shows that there are some similarities in factors affecting the investments in industrial properties between municipalities. Investing in industrial properties in an A-location appears to be linked to all levels of the location: micro, neighborhood, and the attractiveness of the neighborhood. Furthermore, there was a general attitude that proximity to the highway and high technical standards were positive factors for investing in industrial property. Reasons for investing in a property for residential purposes appear to be strongly linked to how the detailed development plan in the area matches the investor's intention of buying the property, which in short-term doesn’t have anything with the location itself to do, but in the long term can affect the neighborhood. Other factors for investing in a property for residential purposes may be related to the neighborhood where community service and urban environment were considered attractive factors for a location. An interesting difference when it comes to real estate investments with housing development purposes is that the physical factors seem to have a lesser impact for the investor in Stockholm municipality, compared to Täby where the investor considered it very important.

This work deals with methods of web page classification. It explains the concept of classification and different features of web pages used for their classification. Further it analyses representation of a page and in detail describes classification method that deals with hierarchical category model and is able to dynamically create new categories. In the second half it shows implementation of chosen method and describes the results.

This paper presents problem of automatic webpages classification using association rules based classifier. Classification problem is presented, as a one of  datamining technique, in context of mining knowledges from text data. There are many text document classification methods presented with highlighting benefits of classification methods using association rules. The main goal of work is adjusting selected classification method for relation data and design draft of webpages classifier, which classifies pages with the aid of visual properties - independent section layout on the web page, not (only) by textual data. There is also ARC-BC classification method presented as a selected method and as one of intriguing classificators, that derives accuracy and understandableness benefits of all other methods.

In this article we establish a bridge between dynamical systems, including topological and measurable dynamical systems as well as continuous skew product flows and nonautonomous dynamical systems; and coalgebras in categories having all finite products. We introduce a straightforward unifying definition of abstract dynamical system on finite product categories. Furthermore, we prove that such systems are in a unique correspondence with monadic algebras whose signature functor takes products with the time space. We substantiate that the categories of topological spaces, metrisable and uniformisable spaces have exponential objects w.r.t. locally compact Hausdorff, σ-compact or arbitrary time spaces as exponents, respectively. Exploiting the adjunction between taking products and exponential objects, we demonstrate a one-to-one correspondence between monadic algebras (given by dynamical systems) for the left-adjoint functor and comonadic coalgebras for the other. This, finally, provides a new, alternative perspective on dynamical systems.

La notion d’acte plurilatéral est issue d’une doctrine importée en France par Léon Duguit au début du XXe siècle. D’origine allemande, cette doctrine consiste à subsumer tous les accords de volontés sous une méta-catégorie nommée « acte plurilatéral », dont le contrat n’est qu’une sous-catégorie. Avec l’acte unilatéral, l’acte plurilatéral forme la summa divisio des actes juridiques. Comme l’acte unilatéral, l’acte plurilatéral connaît plusieurs subdivisions catégorielles auxquelles sont attachés des régimes juridiques distincts. Face à la crise contractuelle que traverse notre ordre juridique et à la multiplication des pseudo-contrats, le concept d’acte plurilatéral constitue un palliatif nécessaire. Remis au goût du jour, il permet ainsi une classification des accords de volontés à même de clarifier une matière en proie à un certain désordre
The notion of plurilateral decision comes from a doctrine imported in France by Leon Duguit at the beginning of the 20th century. Of German origin, this doctrine consists in subsuming all the agreements under a meta-category called “plurilateral decision”, the contract of which is only a subcategory. Along with the unilateral decision, the plurilateral decision constitutes the summa divisio of legal acts. Like the unilateral decision, the plurilateral decision has several categorial subdivisions to wich distinct legal regimes are attached. Faced with the contractual crisis our legal order is going through and the multiplication of pseudo contracts, the concept of plurilateral decision represents a necessary stopgap. Modernized, it thus provides a classification of agreements able to clarify a subject presently faced to some disorder

This thesis deals with machine learning techniques for the extraction of structure and the analysis of the vertebrate olfactory pathway based on related methods. Some of its main contributions are summarized below. We have performed a systematic investigation for classification in biomedical images with the goal of recognizing a material in these images by its texture. This investigation included (i) different measures for evaluating the importance of image descriptors (features), (ii) methods to select a feature set based on these evaluations, and (iii) classification algorithms. Image features were evaluated according to their estimated relevance for the classification task and their redundancy with other features. For this purpose, we proposed a framework for relevance and redundancy measures and, within this framework, we proposed two new measures. These were the value difference metric and the fit criterion. Both measures performed well in comparison with other previously used ones for evaluating features. We also proposed a Hopfield network as a method for feature selection, which in experiments gave one of the best results relative to other previously used approaches. We proposed a genetic algorithm for clustering and tested it on several realworld datasets. This genetic algorithm was novel in several ways, including (i) the use of intra-cluster distance as additional optimization criterion, (ii) an annealing procedure, and (iii) adaptation of mutation rates. As opposed to many conventional clustering algorithms, our optimization framework allowed us to use different cluster validation measures including those which do not rely on cluster centroids. We demonstrated the use of the clustering algorithm experimentally with several cluster validity measures as optimization criteria. We compared the performance of our clustering algorithm to that of the often-used fuzzy c-means algorithm on several standard machine learning datasets from the University of California/Urvine (UCI) and obtained good results. The organization of representations in the brain has been observed at several stages of processing to spatially decompose input from the environment into features that are somehow relevant from a behavioral or perceptual standpoint. For the perception of smells, the analysis of such an organization, however, is not as straightforward because of the missing metric. Some studies report spatial clusters for several combinations of physico-chemical properties in the olfactory bulb at the level of the glomeruli. We performed a systematic study of representations based on a dataset of activity-related images comprising more than 350 odorants and covering the whole spatial array of the first synaptic level in the olfactory system. We found clustered representations for several physico-chemical properties. We compared the relevance of these properties to activations and estimated the size of the coding zones. The results confirmed and extended previous studies on olfactory coding for physico-chemical properties. Particularly of interest was the spatial progression by carbon chain that we found. We discussed our estimates of relevance and coding size in the context of processing strategies. We think that the results obtained in this study could guide the search into olfactory coding primitives and the understanding of the stimulus space. In a second study on representations in the olfactory bulb, we grouped odorants together by perceptual categories, such as floral and fruity. By the application of the same statistical methods as in the previous study, we found clustered zones for these categories. Furthermore, we found that distances between spatial representations were related to perceptual differences in humans as reported in the literature. This was possibly the first time that such an analysis had been done. Apart from pointing towards a spatial decomposition by perceptual dimensions, results indicate that distance relationships between representations could be perceptually meaningful. In a third study, we modeled axon convergence from olfactory receptor neurons to the olfactory bulb. Sensory neurons were stimulated by a set of biologically-relevant odors, which were described by a set of physico-chemical properties that covaried with the neural and glomerular population activity in the olfactory bulb. Convergence was mediated by the covariance between olfactory neurons. In our model, we could replicate the formation of glomeruli and concentration coding as reported in the literature, and further, we found that the spatial relationships between representational zones resulting from our model correlated with reported perceptual differences between odor categories. This shows that natural statistics, including similarity of physico-chemical structure of odorants, can give rise to an ordered arrangement of representations at the olfactory bulb level where the distances between representations are perceptually relevant.

QC 20120224

La sélection de modèle est un thème majeur de l'apprentissage statistique. Dans ce manuscrit, nous introduisons des méthodes de sélection de modèle dédiées à des SVM bi-classes et multi-classes. Ces machines ont pour point commun d'être à coût quadratique, c'est-à-dire que le terme empirique de la fonction objectif de leur problème d'apprentissage est une forme quadratique. Pour les SVM, la sélection de modèle consiste à déterminer la valeur optimale du coefficient de régularisation et à choisir un noyau approprié (ou les valeurs de ses paramètres). Les méthodes que nous proposons combinent des techniques de parcours du chemin de régularisation avec de nouveaux critères de sélection. La thèse s'articule autour de trois contributions principales. La première est une méthode de sélection de modèle par parcours du chemin de régularisation dédiée à la l2-SVM. Nous introduisons à cette occasion de nouvelles approximations de l'erreur en généralisation. Notre deuxième contribution principale est une extension de la première au cas multi-classe, plus précisément à la M-SVM². Cette étude nous a conduits à introduire une nouvelle M-SVM, la M-SVM des moindres carrés. Nous présentons également de nouveaux critères de sélection de modèle pour la M-SVM de Lee, Lin et Wahba à marge dure (et donc la M-SVM²) : un majorant de l'erreur de validation croisée leave-one-out et des approximations de cette erreur. La troisième contribution principale porte sur l'optimisation des valeurs des paramètres du noyau. Notre méthode se fonde sur le principe de maximisation de l'alignement noyau/cible, dans sa version centrée. Elle l'étend à travers l'introduction d'un terme de régularisation. Les évaluations expérimentales de l'ensemble des méthodes développées s'appuient sur des benchmarks fréquemment utilisés dans la littérature, des jeux de données jouet et des jeux de données associés à des problèmes du monde réel
Model selection is of major interest in statistical learning. In this document, we introduce model selection methods for bi-class and multi-class support vector machines. We focus on quadratic loss machines, i.e., machines for which the empirical term of the objective function of the learning problem is a quadratic form. For SVMs, model selection consists in finding the optimal value of the regularization coefficient and choosing an appropriate kernel (or the values of its parameters). The proposed methods use path-following techniques in combination with new model selection criteria. This document is structured around three main contributions. The first one is a method performing model selection through the use of the regularization path for the l2-SVM. In this framework, we introduce new approximations of the generalization error. The second main contribution is the extension of the first one to the multi-category setting, more precisely the M-SVM². This study led us to derive a new M-SVM, the least squares M-SVM. Additionally, we present new model selection criteria for the M-SVM introduced by Lee, Lin and Wahba (and thus the M-SVM²). The third main contribution deals with the optimization of the values of the kernel parameters. Our method makes use of the principle of kernel-target alignment with centered kernels. It extends it through the introduction of a regularization term. Experimental validation of these methods was performed on classical benchmark data, toy data and real-world data

Die seit 2004 in Deutschland bekannt gewordenen Fälle der illegalen Rückführung und irrtümlichen Fehlverbringung von gemäß der Verordnung (EG) Nr. 1774/2002 nicht für den Genuss durch den Menschen bestimmten tierischen Nebenprodukten (TNP) der Kategorie 3 in die Lebensmittelkette haben zu der politischen Diskussion beigetragen, ob die Pflicht der Materialidentifizierbarkeit durch das Getrennthalten TNP am Ort des Anfalls sowie die ausschließliche Kennzeichnung ihrer Transportbehälter bei der Beförderung einen ausreichenden Schutz der Verbraucher garantieren können. Um eine ordnungsgemäße Verwendung TNP der Kategorie 3 sicherzustellen, hat der Bundesrat ihre unmittelbare und eindeutige Kennzeichnung, z.B. durch Farbstoffe, gefordert. Ziel dieser Arbeit war es, einen geeigneten, futtermittelrechtlich zugelassenen Marker für Schlachtnebenprodukte der Kategorie 3 zu erörtern, der eine technisch praktikable, vom Ort des Anfalls bis zum Verarbeitungsbetrieb optisch eindeutige, dauerhafte und nach der Verarbeitung nachweisbare sowie umwelt- und wirtschaftsverträgliche Markierung von Schlachtnebenprodukten zur sicheren Verfolgbarkeit ihrer bestimmungsgemäßen Verwendung als TNP der Kategorie 3 ermöglicht, um ihren Eintrag in die Lebensmittelkette zu unterbinden, ohne die Neutralität der Endprodukte bei der Verwendung markierter TNP als Rohstoffe für Futtermittel zu beeinträchtigen. Für die Markierung von Schlachtnebenprodukten mittels Sprühsystemen wurden für Futtermittel zugelassene, färbende Zusatzstoffe (Verordnung (EG) Nr. 1831/2003) sowie in der medizinischen Diagnostik etablierte Fluoreszenz-Farbstoffe ausgewählt und hinsichtlich der Eindeutigkeit ihrer Markierung, ihrer Farbhaltung nach Bearbeitung sowie ihrer optischen Neutralität in Lebens- und Futtermitteln, die aus markierten TNP hergestellt worden sind, von fünf ungeschulten Prüfpersonen im Rahmen einer einfach beschreibenden, sensorischen und unabhängigen Prüfung gemäß §35 LMBG (L 00.90-6, ASU) beurteilt. Die Ergebnisse der sensorischen Prüfung wurden mit den RGB-Farbprofilen der markierten und nicht markierten TNP vergleichend analysiert. Zum Nachweis des irrtümlichen oder vorsätzlichen Eintrags von mit den ausgewählten Markerfarbstoffen markierten TNP in Lebensmitteln konnten die Analyseverfahren Dünnschichtchromatographie (DC), optische Emissionsspektrometrie mit induktiv gekoppeltem Hochfrequenzplasma (ICP-OES), Photometrie sowie die Fluoreszenzspektrometrie hinsichtlich ihrer Anwendbarkeit untersucht werden. Die Untersuchung der sensorischen Neutralität der Markerfarbstoffe im Endprodukt Futtermittel erfolgte unter anderem durch einen Futtermittelpräferenzversuch an neun Hunden der Rasse Beagle. Brillantsäuregrün E142 (1,3 mg E142/kg TNP) konnte auf Grund der Eindeutigkeit der Markierung von Schlachtnebenprodukten durch die gegenüber den nativen TNP signifikant unterschiedlichen Rot-Farbintensitäten bei gleichzeitiger Neutralität in den Endprodukten (Lebens- oder Futtermittel) und einer guten bis sehr guten Farbhaltung nach dem Waschen der Nebenprodukte, der Kühl- (8°C über zwei Tage) sowie Gefrierlagerung (-25°C über 14 Tage) und dem Verwenden einer 14, 90 als auch 150 Tage gelagerten Farbstofflösung in Kombination mit dem chemisch nachweisbaren Titandioxid (90 mg E171/ kg TNP) als Markerfarbstofflösung zur eindeutigen Markierung von Nebenprodukten der Schlachtung am Ort ihres Anfalls selektiert werden. Die für die Markierung bestimmten Dosierungen der Markerfarbstoffe gelten für Tiere und Menschen als unbedenklich. Die mit den färbenden Zusatzstoffen E142 und E171 markierten Nebenprodukte der Schlachtung können mittels DC (Nachweisgrenze: ≥7,5 µg E142/kg Probe) beziehungsweise ICP-OES und Photometrie (Nachweisgrenze ICP-OES: 8,3 mg E171/kg Probe) ab einem eingebrachten Anteil von 0,55% (DC: E142) beziehungsweise 9% (ICP-OES: E171) in diversen Produkten (Lebens- oder Futtermittel) nachgewiesen werden. In den chemisch und thermisch extrahierten Fetten aus markierten, fettreichen TNP waren die Farbstoffe E142 und E171 jedoch nicht nachweisbar. Eine Fluoreszenzmarkierung TNP kann hingegen nicht präferiert werden, da nicht markierte Nebenprodukte der Schlachtung eine sichtbare und fluoreszenzspektrometrisch nachweisbare Autofluoreszenz aufweisen und in den thermisch verarbeiteten Produkten keine für die Fluoreszenz-farbstoffe charakteristischen Absorptions- und Emissionsspektren nachweisbar waren. Die Markierung mittels Sprühtechnik erscheint unter den Aspekten Substanzverlust und adaptierter Markerfarbstoff pro Kilogramm TNP praktikabel. Die im Labor bestimmte Markierungszeit für TNP (5 sec./kg) ist unter Einbeziehung der Durchsatzraten am Schlachthof als zu lang zu bewerten. Durch die rückstandsfreie Entfernung der Farbstoffe von Edelstahl- und glatten Kunststoffflächen sowie glasierten Fliesen ergeben sich keine Nachteile der Markierung TNP für die Produktion von Futtermitteln und technischen Erzeugnissen. Die in dem Präferenzversuch untersuchten Futtermittel für Hunde aus markierten TNP zeigten keine Abweichungen von der handelsüblichen sensorischen Produktqualität und hinsichtlich ihrer Haltbarkeit durch Sterilisation (F0-Wert). Mit E142 und E171 markierte TNP (Kat. 3) eignen sich somit als Rohstoffe zur Herstellung von Heimtierfuttermitteln. Bei Anwendung einer Kombinationsfarbstofflösung (E142 und E171) würden die für die Marker anfallenden Kosten pro Tonne TNP bis zu 33 Euro betragen. Bei der ausschließlichen Verwendung von E142, welches der optisch eindeutig markierende Farbstoff ist und das eine hohe Sensitivität im dünnschichtchromatographischen Nachweis zeigt, würden die Kosten 1,70 bis 3,40 Euro/t betragen. Bisher konnte kein EU-einheitlicher Rechtsrahmen zur Markierung TNP der Kategorie 3 gestaltet werden. Die politische Diskussion wird aber vor allem national fortgesetzt
Since 2004 several illegal or aberrant transfers of animal by-products (ABP) from category 3 (according to Regulation (EC) No. 1774/2002: not intended for food production) back into food chain, have led to the political discussion, whether duty of material identifiability by separate storing of ABP on site and sole labeling of containers during transport are sufficient to protect consumers from ABP not intended for human consumption. To guarantee adequate utilisation of ABP from category 3, the German Federal Council claimed for an immediate and conclusive marking of ABP by dyeing or similar solutions. This study was implemented to define a convenient, registered feed additive for dyeing of slaughter by-products from category 3, which realize a feasible, from extraction to processing visually conclusive, long-lasting, traceable as well as sustainable and cost-effective marking on site to ensure traceability of intended utilisation as ABP from category 3 and to prevent their influx into food chain, without an impairment of the neutrality of products (e.g. pet food) made from marked ABP. For marking of slaughter by-products by air spraying device, registered colouring feed additives (Regulation (EC) No. 1831/2003) as well as diagnostically established fluorescence pigments were selected and investigated regarding their marking unambiguousness, colour retention after processing and visual neutrality in food and feed made from marked ABP by evaluation of five untrained judging persons in the course of a simply delineative, sensorial and impartial test (official list of analysis methods, ASU §35 LMBG, L 00.90-6), and by comparative RGB-colour measurement of images scanned from stained ABP samples. Detection of aberrant or deliberate discharge of marked ABP into food production was evaluated by investigation of thin layer chromatography (TLC), inductively coupled plasma optical emission spectrometry (ICP-OES), photometry and fluorescence spectrometry. Neutrality of marking feed additives in feedstuff was determined by a feeding preference test with nine dogs of the Beagle breed using pet food made from unmarked und marked ABP. Lissamine Green E142 (1,3 mg E142 per kg ABP) was selected as marker dye for slaughter by-products on site based on its unambiguousness of marking due to the significant different red-colour intensity compared to the non-marked ABP as well as the simultaneous neutrality of the colouring additive E142 in the final products feed and food. Colour retention of E142 marking was conclusive with regard to handling by washing, cold (8°C for two days) respectively fridge storage (-25°C for 14 days) and utilisation of a 14-, 90- and 150-days-stored marker solution. For marking, Lissamine Green was combined with the chemical detectable and registered food colour titanium dioxide (E171: 90 mg/kg ABP). The marker additives are classified as safe for humans and animals within the preferred concentrations for colouring ABP. With E142 und E171 marked ABP were traceable in food and feed using detection methods TLC (limit of detection: ≥7,5 µg E142 per kg sample), photometry and ICP-OES (limit of detection: ≥8,3 mg E171 per kg sample) at a proportion of 0,55% (TLC: E142) respectively 9% (ICP-OES: E171), whereas the named markers were not detectable in chemical and thermal extracted fats produced from marked high-fat ABP. Based on the visible and fluorescence spectrometric detectable autofluorescence of animal tissues as well as the uncharacteristic emission and absorption spectra of fluorescence pigments in processed ABP, fluorescence markers are not preferential for marking of slaughter by-products from category 3. Marking of slaughter by-products by air spraying device appeared practicable in due consideration of marker depletion and tissue-adapted marker per kg ABP. Current time of marking under laboratory conditions (5 sec. per kg ABP) must be graded as too long, regarding high transfer rates in slaughterhouses. Concerning the residue-free cleaning of stainless steel and even plastic surfaces from the marker solution, the utilisation of marked ABP for manufacturing of feed and technical products is unproblematic. Investigated pet food samples produced from marked ABP were from comparable commercial sensory product quality and showed no deviation of normal storability due to sterilisation. In conclusion, with E142 and E171 visible marked ABP from category 3 are suitable as crude materials for pet food production. The application of the combined marker solution (E142 and E171) have to be evaluated as comparative expensive (33 Euro per ton ABP), while the exclusive application of E142 as the optic conclusive and sensitive detectable marker for ABP is associated with sustainable costs from 1,70 to 3,40 Euro per ton ABP. To date, an EU-common regulatory framework for marking of ABP from category 3 could not be specified. Nevertheless the political discussion is still continued, especially in Germany

My diploma thesis deals with a connection of Videňská street to the future residential district situated between Přízřenice and Modřice. The level of this thesis is a technical study. The main topic of the thesis is a design of an interchange on the Vídeňská and Moravanská streets. The thesis is focused on a technical verification of variable ramps’ connections from Vídeňská street to Moravanská street. Another important point of the thesis is an improvement of security quality of cycle traffic and a creation of a traffic junction of the city transportation system in this area. The SWOT analysis was applied to evaluation and selection the most advantageous variant of the variants. Depending on SWOT analysis, the thesis is focused on more detailed design of the most advantageous variant.

In this paper, we generalize the notions of polymorphisms and invariant relations to arbitrary categories. This leads us to a Galois connection that coincides with the classical case from universal algebra if the underlying category is the category of sets, but remains applicable no matter how the category is changed. In analogy to the situation in universal algebra, we characterize the Galois closed classes by local closures of clones of operations and local closures of what we will introduce as clones of (generalized) relations. Since the approach is built on purely category-theoretic properties, we will also discuss the dualization of our notions.

Die Dissertation besteht aus einem Einleitungsteil und drei empirischen Beiträgen. Die Einleitung gibt eine Einführung in das Themengebiet der Selbstwertschätzung und deren Erfassung mit indirekten Verfahren. Vor allem aber dient sie dazu, die empirischen Befunde der Beiträge zu integrieren. Die ersten beiden Artikel widmen sich der Überprüfung der psychometrischen Qualität indirekter Verfahren. Das Augenmerk des ersten Beitrags liegt auf der Prüfung, ob verschiedene indirekte Verfahren konvergieren und ob explizite und implizite Selbstwertschätzung korrespondieren, wenn implizite Selbstwertschätzung reliabel erfasst wird. Außerdem werden strukturelle Ähnlichkeiten der indirekten Verfahren in die Betrachtungen mit einbezogen. Der zweite Beitrag komplettiert die Untersuchungen zur Güte der Verfahren. Der Fokus dieses Beitrags liegt im Nachweis der prädiktiven Validität verschiedener indirekter Verfahren. Dabei werden Kriterien herangezogen, die nicht im Selbstbericht erhoben werden. Im Detail wird eine doppelte Dissoziation zwischen expliziter und impliziter Selbstwertschätzung geprüft. Der dritte Beitrag rundet diese Arbeit ab, da er sich auch einem anwendungsbezogenen Aspekt der Persönlichkeitspsychologie widmet. Untersucht wird der Zusammenhang zwischen defensiven Verhaltensweisen und Selbstwertdiskrepanzen, also der Kombination expliziter und impliziter Selbstwertschätzung
The dissertation consists of an introduction and three empirical journal articles. The introduction gives the theoretical background about self-esteem and its assessment with indirect measures, and primarily, integrates the three journal articles. The first article investigates the reliability and convergent and discriminant validity of indirect measures of self-esteem. The second article complements the examination of the psychometric properties. It contributes evidence to the predictive validity of indirect self-esteem measures. With using non-self-reported criteria, a double dissociation between explicit and implicit self-esteem is tested. The third article brings into focus an applied aspect of personality psychology. It is analyzed how defensive reactions are related to self-esteem discrepancies, that are different combinations of explicit and implicit self-esteem

In this thesis, we generalize clones (as well as their relational counterparts and the relationship between them) to categories. Based on this framework, we introduce a general duality theory for clones and apply it to obtain new results for clones on finite sets.

Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2006.
Source: Dissertation Abstracts International, Volume: 67-07, Section: B, page: 3926. Adviser: Dan Roth. Includes bibliographical references (leaves 114-119) Available on microfilm from Pro Quest Information and Learning.

碩士
國立中央大學
資訊管理研究所
98
The news classification problem is concerned with how to assign the correct category for the unclassified news. Although a large number of past studies have studied this problem, a common weakness of these studied is that their classification algorithms were usually designed from technical perspective and they seldom considered how experts really classify the news in a practical classification process. In this research, we first observe how media workers classify news in their daily operations, and we find that their classification process mainly consists of the following operations. (1) If some important keywords or phrases are present in the news, then they directly assign the news to certain categories. (2) Otherwise, they must check in details the whole content of news to determine which category it should belong to. (3) Since a news category may contain several independent but related subcategories, the news is usually classified by assigning it to the most appropriate subcategory, which can in turn determine its category. 　　By imitating the above working process, we proposed a news classification algorithm. In the learning phase, we use associative classification rules to find representative keywords in each category. In addition, we further generate a number of subcategories by clustering news under each category. In the classification phase, we assign unclassified news the most appropriate category by using associative classification rules if rules’ confidence is high enough. Otherwise, we will determine the category by measuring the similarity between unclassified news and subcategories. The experimental comparison shows that our approach has better and more stable classification performance than traditional algorithms.

碩士
國立清華大學
資訊系統與應用研究所
104
In recent years, researches of aspect-category-based sentiment analysis have been approached in terms of predefined categories. In this paper, we target two sub-tasks of SemEval-2014 Task 4 dedicated to aspect-based sentiment analysis: detecting aspect category and aspect category polarity. Also, a pre-identified set of aspect categories {food, price, service, ambience, miscellaneous} defined by SemEval-2014 have been used in this paper. The majority of the submissions worked on these two sub-tasks with machine learning mainly with n-grams and sentiment lexicon features. The difficulty for these submissions is that some opinion word (e.g., "good") is general and cannot be referred to any particular category. By contrast, we use aspect-opinion pairs as one of the features in this paper to overcome this difficulty. To detect these pairs, we identify the opinion words in customer reviews, and then detect their related aspect terms by dependency rule. This system has been done on restaurant domain applying to Chinese customer reviews. Our experiment achieved 87.5% of accuracy by using Word2Vec to detect aspect category polarity. Aspect-opinion pair features employed in this system contribute to 88.3% of accuracy. When all features are employed, the accuracy is improved from 84.4% to 89.0%. Experimental results demonstrate the effectiveness of aspect-opinion pair features applied to the aspect-category-based sentiment classification system.

碩士
淡江大學
資訊工程學系碩士班
104
Rapid and vigorous development of information network technology has resulted in the largest data repository. Collecting relevant information in such a large body of data is rather difficult for any user. This paper is aimed to help users to grasp key information in a short period of time. We observe that term frequency in a article can be used as keyword for that article. Article theme can be easily grasped based on these keywords. Therefore, users can find the information they want through keyword and significantly reduce unnecessary search time. Proper word segmentation enables article theme extraction. And article classification can be achieved by theme differentiation. We use 320 articles in the theme classification experiment. These articles are divided into two categories: training and testing. There are 285 training samples, all belonging to the sports news theme. There are 15 testing samples that are consists of themes picked at random. The result is able to pick out 6 articles which belonging to sport news theme among the 15 testing samples. Among the 20 negative samples, there are 4 false positives, all due to names related to sports events.

碩士
國立交通大學
電機資訊學院碩士在職專班
94
“Directory style” is the most common website data mode. Take shopping websites, for example; classifying products by catalogue can promote the efficiency of product data processing. The problem of product matching is one of very fundamental problems of the website data processing and the purpose of the product classification is to find out which products are belonged to the same kind of products from two websites. Products which are under the related category are possible to be the same kind of products because of product between two websites. Hence, finding out the similarity of the category between websites can elevate the accuracy and efficiency of product matching. This thesis comes up with a kind of improvement method about the categorical correspondence. The traditional VSM (Vector Space Model) can not compare the similarity of two websites accurately. This is because product amount is too few or the product title is inconsistent under the category and will make the similarity of categories be totally different from VSM (Vector Space Model). In order to improve this problem, the thesis proposes a concept of Middle Category which is a catalogue website has general characteristics. Through categorical matching with Middle Category, the accuracy of categorical matching of other websites will increase. This thesis discusses characteristics which are necessary for being Middle Category. In addition, the thesis also proves that similarity of categories will increase by the information of Middle Category. The result of the thesis’ experiment manifests that after categories of websites do the similarity matching by Middle Category; the same categories utilize LMCS (Longest and Most Common Segments) to do product matching that increases recall, precision, and similarity over eighty eight percent.

碩士
國立雲林科技大學
資訊管理系
102
Data mining can use the information and pattern that hidden at the data, they can Classification, prediction, clustering and association. Formal Concept Analysis (FCA) is a data analysis method that found the structure of the theoretical concepts from the data set, it can be formalized into this article tabular graphical concept lattice, and applied on the data’s classification. This study applies the Formal Concept Analysis to classificate the data set of the categorical-type and multivariate attribute. Since the time complexity of FCA in the worst case achieve O(2N) (N is the number of nodess), and the node number is affected by the object number and attribute number. Therefore, this study use record reduction to decrease the object number and use dimension reduction to decrease the attribute number, so that the original high time complexity data set can be classificated, and use system experiments to evaluate the effectiveness. Experimental results show that when using record reduction, the max usable record number (object number) is affected by the attributes number, and their correlation is negative；When using dimensions reduction, the max usable attribute is affected by object number, and their correlation is negative. When data set be processed by record reduction, the data record number will affect the classification rate, and their correlation is positive；When data set be processed by dimensions record reduc, the data attribute number will affect the classification rate, and their correlation is positive. Futhermmore, the dimensions reduction method is better than record reduction method in the classification rate.

碩士
明志科技大學
工業設計系碩士班
102
Successful products are the foundation of any successful business, and the added sales may ensure the sustainability of a company, so product design and development is the foundation of any successful business. A consumer purchase products for their added functional and psychological values, thus the product design must take consumers’ emotional needs into consideration; the products’ added value also comes from innovation, thus the designers cannot afford being limited by their personal knowledge and experiences. Quick to market at reduced time and cost is the Nirvana of product design. This research combines Kansei Engineering and the Knowledge Effects to propose a Structured Innovative Product Design Process. Specifically, the category classification of Kansei Engineering is applied to ensure the proper extraction of consumer emotional needs, and Knowledge Effects database of TRIZ is applied to enhance the designer’s efficiency and effectiveness in converting these needs to design concepts. The applicability and operability of this proposed design process was tested in this research using graduate students enrolled in Ming Chi University of Technology’s Industrial Design Department as well as an experienced practitioner (also an MCUT alumnus). The results indicate that (1) all participants were able to develop satisfactory design concepts using the proposed design process without major difficulty; (2) the effects solutions generated with the help of Knowledge Effects database were rich in variety, which did help designers expand their horizon; (3) participants could easily follow the proposed design process without specific background in TRIZ or Kansei Engineering; and (4) the general-purpose path map developed by the researcher could help further streamline the process.

Dissertação de Mestrado em Engenharia Informática apresentada à Faculdade de Ciências e Tecnologia
Atualmente, os Sistemas de Recomendação têm apresentado um aumento na sua utilização, uma vez que permitem uma melhor entrega de um serviço aos seus utilizadores. No entanto, de forma a que estes sistemas consigam ser robustos e eficientes, é necessária a existência de uma base de conteúdo corretamente classificado. No caso de classificação de programas de televisão, a sua classificação pode, para além de melhorar o conteúdo mostrado no Guia de Programas Eletrónicos, melhorar a base de dados utilizada para os Sistemas de Recomendação e consequentemente o sistema que faz uso desses dados. De forma a automatizar o processo, o Processamento de Linguagem Natural surge como um conjunto de técnicas muito útil, uma vez que permite às máquinas entender e processar linguagem natural, assim como classificar grandes quantidades de dados automaticamente. Este documento, elaborado em contexto de Dissertação para obtenção de Mestrado e estágio na empresa Altice Labs em articulação com a Altice/MEO, apresenta o processo de desenvolvimento de um modelo capaz de classificar programas de televisão pela sua categoria, com base no texto da sua sinopse, utilizando técnicas de Processamento de Linguagem Natural. Começamos por discutir o contexto teórico de alguns conceitos importantes para o problema, seguindo-se uma revisão a trabalhos semelhantes de forma aperceber as técnicas tipicamente implementadas e resultados obtidos. Seguidamente, são apresentadas, com detalhe, as tarefas desenvolvidas e decisões tomadas no contexto desta investigação/estágio, às quais acresce uma explicação de como o produto final está a ser integrado nos sistemas da MEO para ser utilizado diretamente nas suas bases de dados. O desenvolvimento de um modelo inteligente para classificação de programas permite à Altice/MEO a automatização dos seus processos de classificação e um consequente melhoramento do conteúdo apresentado no Guia de Programas Eletrónicos (EPG) e nos Sistemas de Recomendação que fazem uso dessa classificação, assim como ajudará também a indicar oportunidades de negócio ao melhorar o detalhe no perfil dos clientes.
Currently, the Recommendation Systems have shown an increase in their use, since they allow better delivery of a service to its users. However, for these systems to be robust and efficient, a correctly classified content base is required. In the case of television program classification, its classification can, besides improving the content shown in the Electronic Program Guide, improve the database used for the Recommendation Systems and consequently the system that makes use of this data. To automate the process, Natural Language Processing appears as a very useful set of techniques, since it allows machines to understand and process natural language, as well as classify large amounts of data automatically. This document, elaborated in the context of a Dissertation to obtain a Master’s degree and internship at the company Altice Labs in articulation with Altice/MEO, presents the process of developing a model capable of classifying television programs by their category, based on the text of their synopsis, using Natural Language Processing techniques. We start by discussing the theoretical context of some important concepts for the problem, followed by a review of similar works to understand the techniques typically implemented and associated results. Next, we present, in detail, the tasks developed and the decision path in the context of this research/internship, as well as how the final product is being integrated into the MEO systems to be used directly on their database. The development of an intelligent model for program classification allows Altice/MEO to automate its classification processes and results in a consequent improvement of both the content presented in the Electronic Program Guide (EPG) and the Recommendation Systems that make use of that classification, as well as helping indicate business opportunities by improving the detail on the client’s profile.

碩士
國立交通大學
工業工程與管理系所
107
Amount and position of defects on TFT-LCD panel affects the level of the panel, adding Automatic optical inspection(AOI) into the light-on test, which can automatically detect defects from the image of LCD panel. This study constructed a model based on Convolutional neural network of deep learning, which can classify various kind of defects at once. Without any pre-processing, the model can automatically capture features through a large number of data and model training, resulting in a high-efficiency and high-accuracy defects classification model. This study used actual panel images from Taiwan's leading computer hardware manufacturers for model construction, model testing and validating the result. After validation, the model constructed by this study has 99.9% model accuracy and excellent specificity and sensitivity, the model can also finish the process of classifying a TFT-LCD panel in only 467 seconds.

Thyroid nodules are a common entity, the majority being benign. Thyroid malignancies, however, present themselves as nodules in 95% of the cases. “Fine needle aspiration plays an essential role in the evaluation of the euthyroid patient with a thyroid nodule”. The Bethesda System for Reporting Thyroid Cytopathology is an international classification of cytologic findings that links to clinical management. Its follicular lesion of undetermined significance category is employed when “cytomorphologic findings are neither convincingly benign nor sufficient for interpretation of neoplasia/malignancy”. This category proves a challenge in the management of these nodules. Several studies are ongoing to determine the diseases underlying a follicular lesion of undetermined significance category report. In this study a correlation between this cytological cathegory and a histological diagnosis is made using data from Portuguese hospitals. The percentage of this cytological cathegory and those who underwent surgery was calculated. The post surgical histological diagnosis was noted and their relative percentage calculated. The consulted hospitals had 5267 thyroid cytology reports and 310 were reported as follicular lesion of undetermined significance. A hundred and one underwent surgery and the histological report was consulted. Each histological diagnosis was noted individually and because many cases had more than one histological diagnosis a single dominant disease was considered for statistical study. Four groups where created according to the dominant diagnosis: malignant neoplasia, benign neoplasia, inflammation and hyperplasia. A fifth group was created for diseases that had an uncertain malignant potential and for micro carcinomas. Since inflammation can be a stimulus to malignant disease, all histological diagnosis associated with inflammation where noted. The malignant neoplasia group had 28 cases, the benign neoplasia group had 25 cases, the inflammation group had 10 cases and the hyperplasia group had 30 cases. Micro carcinomas and tumors with uncertain malignant potential had 8 cases in total. Inflammation was present in 22 cases. This study has found that most cases associated with FLUS are hyperplasic in origin. Malignancies where higher than expected, being the predominant disease in 27.72% of all cases. The total cases of benign and malignant neoplasias combined far outnumber the remainder cases and FNAC should appropriately triage surgical cases. Inflammation coexisted with malignant and benign neoplasias and with microcarcinomas and cases in the inflammatory group had some form of hyperplasia. With these findings it is pertinent to question the influence of inflammation in a FLUS diagnosis.
Os nódulos tiroideus são uma entidade comum, sendo a sua maioria benignos. As neoplasias malignas da tiróide apresentam-se, contudo, como nódulos em 95% dos casos. A citologia aspirativa por agulha fina é fundamental para o estudo dos nódulos tiroideus, pois reduz cirurgias nos pacientes com lesões benignas e tria adequadamente pacientes com nódulos malignos ou potencialmente malignos para intervenção cirúrgica. A nomenclatura de Bethesda é uma proposta internacional de classificação dos achados citopatológicos. Uma das suas categorias, a “lesão folicular de significado indeterminado” é utilizada quando os achados citomorfológicos não são nem convincentemente benignos nem suficientes para uma interpretação de malignidade. Nódulos com esta classificação revelam-se um desafio em termos de manejo clínico. Vários estudos têm sido feitos para determinar que patologias estão na origem de uma lesão folicular de significado indeterminado. Neste estudo é feita uma correlação cito-histológica dos casos de lesão folicular de significado indeterminado submetidos a cirurgia, utilizando dados do Instituto Português de Oncologia Francisco Gentil de Lisboa, do Hospital da Luz e do Hospital Beatriz Ângelo. Os hospitais consultados tinham 5267 relatórios de citologias tiroideias entre 1 de Janeiro de 2012 e 31 de Dezembro de 2013 e 310 foram de classificados como “lesão folicular de significado indeterminado”. Cento e um casos foram submetidos a cirurgia e o relatório anátomo-patológico foi consultado. Cada diagnóstico histológico foi anotado individualmente e como muitos casos tinham mais do que uma patologia um único diagnóstico dominante foi considerado para o estudo estatístico. Quatro grupos foram criados de acordo com um diagnóstico dominante: neoplasia maligna, neoplasia benigna, inflamação e hiperplasia. Um quinto grupo foi criado para doenças com potencial maligno incerto e para micro carcinomas. Uma vez que a inflamação pode ser um estímulo para a malignidade, todos diagnóstico histológico associados a inflamação foram inventariados. As neoplasias malignas foram o diagnóstico dominante em 28 casos, correspondendo a 27.72% de todos os casos. As neoplasias benignas foram o diagnóstico dominante em 25 casos e a inflamação em 10 dos casos. A hiperplasia foi considerada como diagnóstico dominante em 30 dos casos, sendo este o maior grupo. Os micro carcinomas e os tumores com potencial maligno incerto perfizeram 8 casos no total. A inflamação estava presente em 22 casos. Na população estudada revelou-se que a maioria dos casos de lesão folicular de significado indeterminado tinham o diagnóstico histológico de hiperplasia. As neoplasias malignas tiveram uma frequência de 27.72%, maior que o esperado. O total de casos de neoplasias benignas e malignas corresponderam a mais de metade de todos os casos e a citologia aspirativa de agulha fina deveria ter triado adequadamente estes casos.

Research on machine learning has taken numerous different directions. The present study focussed on the microstructural characteristics of learning systems. It was postulated that learning systems consist of a macrostructure which controls the flow of information, and a micro-structure which manipulates information for decision making. A review of the literature suggested that the basic function of the micro-structure of learning systems was to make a choice among a set of alternatives. This decision function was then equated with the task of making classification decisions. On the basis of the requirements for practical learning systems, the feature frequency approach was chosen for model development. An analysis of the feature frequency approach indicated that an effective model must be sensitive to both within-dimension and between-category variations in frequencies. A model was then developed to provide for such sensitivities. The model was based on the Bayes' Theorem with an assumption of uniform prior probability of occurrence for the categories. This model was tested using data collected for neuropsychological diagnosis of children. Results of the tests showed that the model was capable of learning and provided a satisfactory level of performance. The performance of the model was compared with that of other models designed for the same purpose. The other models included NEXSYS, a rule-based system specially design for this type of diagnosis, discriminant analysis, which is a statistical technique widely used for pattern recognition, and neural networks, which attempt to simulate the neural activities of the brain. Results of the tests showed that the model's performance was comparable to that of the other models. Further analysis indicated that the model has certain advantages in that it has a simple structure, is capable of explaining its decisions, and is more efficient than the other models.
Graduate

博士
國立臺灣大學
圖書資訊學研究所
104
The Impact Factor (IF) ranking of journals in Journal Citation Reports (JCR) influences journal’s academic evaluation and the IF ranking is close related to its journal category. Therefore, I put this research into the perspective of academic evaluation to analyze the 88 “Information Science and Library Science” (IS&LS) journals that are included in JCR to explore the category problem of JCR through bibliometrics. I analyze the citations and author’s institutions of the four different sub-categories, “LS” (Library Science)、”IS” (Information Science)、”SM” (Scientometrics) and “MIS” (Management Information System), and four specific journals (LISR、JASIST、SMs、MISQ), also examine the Impact Factor and Quartile distribution of these journals. As for discipline of citation, my research shows that the articles in IS&LS、LS、IS and SM mostly cite the literature from the LIS, except articles in MIS. Articles in LS, IS and SM cite the literature in LIS, computer science, general science and medicine most, while the articles in MIS most cite the literature from computer science, MIS, management and business. It worth mentioned that the citation disciplines between LIS and MIS are not very close. As for citation in specific journal, LISR, JASIST and SMs are mostly citing and cited by journals in LIS and are least by journals in MIS. MISQ is mostly citing and cited by journals in computer science and business, has no cross citation with journals in LIS, which means the selected journal between MIS and LIS have few cross citations. As for the author’s institutions, authors in IS&LS, LIS journals are mostly based in LIS institutions; authors in MIS basically come from business school and MIS, few come from LIS. It is obvious that the authors in MIS and LIS journals do not come from the similar discipline institutions. The authors in selected journals also show the similar feature. The authors in my selected journal of LIS are from LIS; authors in MISQ are mostly from business school, MIS and management, very few come from LIS. My research finds that the citation discipline and author’s institution discipline of journals in four subcategories are with high heterogeneity in MIS and LIS (LS, IS, SM). MIS is inclined to transdisciplinary and has faster development than LIS. The IF is significantly different between MIS and LIS. The Q value of MIS falls in Q1, while the Q value of LS mostly falls in Q4. LIS though leads in numbers of journals, articles, and authors but the average numbers of citation per article are far behind the MIS, which means the difference between LIS and MIS falls not only in citation discipline and author background but also in citation behaviors. According to the citation analysis in this research, my finding shows that IS&LS and LIS are statically significant in citation discipline and author’s institutions. However, MIS and LIS have significant difference in citation discipline, author institutions and IF. As we can see, the top 25% in journal ranking are mostly occupied by MIS journals, which might have negative impact on the IF ranking and academic evaluation of the LIS journals and authors. Besides, LIS journals seldom cite the MIS journals, and articles in MIS almost don’t cite articles in LIS Journals. The authors in MIS and LIS journals mostly come from institutions in different disciplines. To conclude, the journals in MIS and LIS are heterogeneous.

Diese Arbeit beschäftigt sich mit der Speziestheorie aus Mochizukis Beweis(versuch) der abc-Vermutung. Es wird ein Standpunkt eingeführt, der Parallelen zwischen der Kategorientheorie und der Speziestheorie aufzeigt und es werden so die Besonderheiten der Speziestheorie herausgearbeitet. In der Speziestheorie möchte man Konstruktionen ausführen, welche von keinen Auswahlen abhängen. Dieses Problem wird in einem allgemeinen Kontext für universelle Morphismen gelöst. An Beispielen wird die in der Arbeit behandelte Theorie erklärt.

Following the classical theory of Baire category results for sets of measure-preserving transformations, this work develops a theory for Baire category results for sets of measure-preserving extensions. First the case is considered where a measure space and a sub-algebra are fixed, and extensions are considered to be any measure-preserving transformations which leave this sub-algebra invariant. In the latter case, extensions of a fixed measure-preserving transformation are considered. In both cases, it is shown that the set of weakly mixing extensions form a dense, G-delta set

In this work we introduce some category-theoretical concepts and techniques to study probability distributions on metric spaces and ordered metric spaces. In Chapter 1 we give an overview of the concept of a probability monad, first defined by Giry. Probability monads can be interpreted as a categorical tool to talk about random elements of a space X. We can consider these random elements as formal convex combinations, or mixtures, of elements of X. Spaces where the convex combinations can be actually evaluated are called algebras of the probability monad. In Chapter 2 we define a probability monad on the category of complete metric spaces and 1-Lipschitz maps called the Kantorovich monad, extending a previous construction due to van Breugel. This monad assigns to each complete metric space X its Wasserstein space PX. It is well-known that finitely supported probability measures with rational coefficients, or empirical distributions of finite sequences, are dense in the Wasserstein space. This density property can be translated into categorical language as a colimit of a diagram involving certain powers of X. The monad structure of P, and in particular the integration map, is uniquely determined by this universal property. We prove that the algebras of the Kantorovich monad are exactly the closed convex subsets of Banach spaces. In Chapter 3 we extend the Kantorovich monad of Chapter 2 to metric spaces equipped with a partial order. The order is inherited by the Wasserstein space, and is called the stochastic order. Differently from most approaches in the literature, we define a compatibility condition of the order with the metric itself, rather then with the topology it induces. We call the spaces with this property L-ordered spaces. On L-ordered spaces, the stochastic order induced on the Wasserstein spaces satisfies itself a form of Kantorovich duality. The Kantorovich monad can be extended to the category of L-ordered metric spaces. We prove that its algebras are the closed convex subsets of ordered Banach spaces, i.e. Banach spaces equipped with a closed cone. The category of L-ordered metric spaces can be considered a 2-category, in which we can describe concave and convex maps categorically as the lax and oplax morphisms of algebras. In Chapter 4 we develop a new categorical formalism to describe operations evaluated partially. We prove that partial evaluations for the Kantorovich monad, or partial expectations, define a closed partial order on the Wasserstein space PA over every algebra A, and that the resulting ordered space is itself an algebra. We prove that, for the Kantorovich monad, these partial expectations correspond to conditional expectations in distribution. Finally, we study the relation between these partial evaluation orders and convex functions. We prove a general duality theorem extending the well-known duality between convex functions and conditional expectations to general ordered Banach spaces.

In Scandinavian languages there are many popular bird names, most of which have been recorded in the dialect archives in the Nordic countries. The thesis concentrates on the bird names for snipes, i.e. the common snipe, the great snipe, the jacksnipe and the woodcook, in all approx. 330 different names. Some of these are recognized throughout Scandinavia, while others are only sparsely confirmed in single dialects. An especially large number of names refer to the common snipe, mainly due to its typical vibrating sound. Since the sound resembles a certain unobtrusive whinny from a horse or bleat from a goat or a sheep, many of the names contain words associating to these animals. Names describing a creaking sound mostly refer to the woodcock. Fairly common are names describing the protruding beak, a characteristic feature of all snipes. In order to give a complete semantic and etymological picture, all words forming part of the names have been identified and sorted by category, i.e. words indicating a common implicit meaning. The categories in turn have been divided into two main groups: words describing different sounds, and words describing visual impressions. In addition, names describing popular beliefs and those transferred from other bird species are presented. Factors that influence origin, formation, existence and development of bird names have been considered. With the etymological discussion forming the necessary background, ethnological aswell as cultural influences complete the picture of the naming process. Man's need for identification and classification of phenomena in his environment as well as the presence vs. lack of affect are important factors in the process. As is shown, a striking sound along with a particular appearence has inspired many affected popular names, while scientific names largely refer to visual, non-affected impressions. Due to man's present lack of everyday connection with nature many of the popular bird names presented are no longer in use. However, some of the names prove to be fairly young, which shows that the process of popular classification and naming is still a functioning part of our language and culture.

Doktorsavhandling vid Uppsala universitet, 1996.