Rozprawy doktorskie na temat „Similarity-based Reasoning”

Continuous development has been occurring in the area of decision support systems. Modern systems focus on applying decision models that can provide intelligent support to the decision maker. These systems focus on modelling the human reasoning process in situations requiring decision. This task may be achieved by using an appropriate decision model. Multicriteria decision making (MCDM) is a common decision making approach. This research investigates and seeks a way to resolve various issues associated with the application of this model. MCDM is a formal and systematic decision making approach that evaluates a given set of alternatives against a given set of criteria. The global evaluation of alternatives is determined through the process of aggregation. It is well established that the aggregation process should consider the importance of criteria while determining the overall worth of an alternative. The importance of individual criteria and of sub-sets of the criteria affects the global evaluation. The aggregation also needs to consider the importance of the sub-set of criteria. Most decision problems involve dependent criteria and the interaction between the criteria needs to be modelled. Traditional aggregation approaches, such as weighted average, do not model the interaction between the criteria. Non-additive measures such as fuzzy measures model the interaction between the criteria. However, determination of non-additive measures in a practical application is problematic. Various approaches have been proposed to resolve the difficulty in acquisition of fuzzy measures. These approaches mainly propose use of past precedents. This research extends this notion and proposes an approach based on similarity-based reasoning. Solutions to the past problems can be used to solve the new decision problems. This is the central idea behind the proposed methodology. The methodology itself applies the theory of reasoning by analogy for solving MCDM problems. This methodology uses a repository of cases of past decision problems. This case base is used to determine the fuzzy measures for the new decision problem. This work also analyses various similarity measures. The illustration of the proposed methodology in a case-based decision support system shows that interactive models are suitable tools for determining fuzzy measures in a given decision problem. This research makes an important contribution by proposing a similarity-based approach for acquisition of fuzzy measures.

Continuous development has been occurring in the area of decision support systems. Modern systems focus on applying decision models that can provide intelligent support to the decision maker. These systems focus on modelling the human reasoning process in situations requiring decision. This task may be achieved by using an appropriate decision model. Multicriteria decision making (MCDM) is a common decision making approach. This research investigates and seeks a way to resolve various issues associated with the application of this model. MCDM is a formal and systematic decision making approach that evaluates a given set of alternatives against a given set of criteria. The global evaluation of alternatives is determined through the process of aggregation. It is well established that the aggregation process should consider the importance of criteria while determining the overall worth of an alternative. The importance of individual criteria and of sub-sets of the criteria affects the global evaluation. The aggregation also needs to consider the importance of the sub-set of criteria. Most decision problems involve dependent criteria and the interaction between the criteria needs to be modelled. Traditional aggregation approaches, such as weighted average, do not model the interaction between the criteria. Non-additive measures such as fuzzy measures model the interaction between the criteria. However, determination of non-additive measures in a practical application is problematic. Various approaches have been proposed to resolve the difficulty in acquisition of fuzzy measures. These approaches mainly propose use of past precedents. This research extends this notion and proposes an approach based on similarity-based reasoning. Solutions to the past problems can be used to solve the new decision problems. This is the central idea behind the proposed methodology. The methodology itself applies the theory of reasoning by analogy for solving MCDM problems. This methodology uses a repository of cases of past decision problems. This case base is used to determine the fuzzy measures for the new decision problem. This work also analyses various similarity measures. The illustration of the proposed methodology in a case-based decision support system shows that interactive models are suitable tools for determining fuzzy measures in a given decision problem. This research makes an important contribution by proposing a similarity-based approach for acquisition of fuzzy measures.
Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
School of Information and Communication Technology
Full Text

Due to the proliferation of object-oriented software development, UML software designs are ubiquitous. The creation of software designs already enjoys wide software support through CASE (Computer-Aided Software Engineering) tools. However, there has been limited application of computer reasoning to software designs in other areas. Yet there is expert knowledge embedded in software design artefacts which could be useful if it were successfully retrieved. Thus, there is a need for automated support for expert knowledge retrieval from software design artefacts. A software design is an abstract representation of a software product and, in the case of a class diagram, contains information about its structure. It is therefore possible to extract knowledge about a software application from its design. For a human expert an important aspect of a class diagram are the semantic tags associated with each composing element, as these provide a link to the concept each element represents. For implemented code, however, the semantic tags have no bearing. The focus of this research has been on the question of whether is it possible to retrieve knowledge from class diagrams in the absence of semantic information. This thesis formulates an approach which combines case-based reasoning with graph matching to retrieve knowledge from class diagrams using only structural information. The practical applicability of this research has been demonstrated in the areas of cost estimation and plagiarism detection. It was shown that by applying case-based reasoning and graph matching to measure similarity between class diagrams it is possible to identify properties of an implementation not encoded within the actual diagram, such as the domain, programming language, quality and implementation cost. An approach for increasing users’ confidence in automatic class diagram matching by providing explanation is also presented. The findings show that the technique applied here can contribute to industry and academia alike in obtaining solutions from class diagrams where semantic information is lacking. The approach presented here, as well as its evaluation, were automated through the development of the UMLSimilator software tool.

This work has investigated the possibility of classifying Java source code into cases for a case-based reasoning system. A Case-Based Reasoning system is a problem solving method in Artificial Intelligence that uses knowledge of previously solved problems to solve new problems. A case in case-based reasoning consists of two parts: the problem part and solution part. The problem part describes a problem that needs to be solved and the solution part describes how this problem was solved. In this work, the problem is described as a Java source file using words that describes the content in the source file and the solution is a classification of the source file along with the source code. To classify Java source code, a classification system was developed. It consists of four analyzers: type filter, documentation analyzer, syntactic analyzer and semantic analyzer. The type filter determines if a Java source file contains a class or interface. The documentation analyzer determines the level of documentation in asource file to see the usefulness of a file. The syntactic analyzer extracts statistics from the source code to be used for similarity, and the semantic analyzer extracts semantics from the source code. The finished classification system is formed as a kd-tree, where the leaf nodes contains the classified source files i.e. the cases. Furthermore, a vocabulary was developed to contain the domain knowledge about the Java language. The resulting kd-tree was found to be imbalanced when tested, as the majority of source files analyzed were placed inthe left-most leaf nodes. The conclusion from this was that using documentation as a part of the classification made the tree imbalanced and thus another way has to be found. This is due to the fact that source code is not documented to such an extent that it would be useful for this purpose.

Schack är ett spel som ofta används för att undersöka olika tekniker inom artificiell intelligens (AI). I det här arbetet ställs frågan om det går att utveckla en AI-agent vars beslutsfattande är baserat på tekniken Case-based Reasoning (CBR) med grundlig likhet som spelar bättre med fallbaser baserade på bättre experter. En AI-agent har utvecklats som spelat ett antal partier mot sig själv med olika fallbaser baserade på olika experter. Efter att ha undersökt resultatet visade de sig att AI-agenten spelar så dåligt att den nästan aldrig lyckades vinna oavsett fallbas, vilket gjorde att det inte gick att rangordna dem efter skicklighet. I framtida arbete är det intressant att undersöka andra likheter än grundlig likhet. Det är även av intresse att undersöka om en CBR-baserad schackspelande AI-agent kan spela schack med hög skicklighet.

This thesis investigates the feasibility of establishing a generalised approach for defining similarity metrics between 3D shapes for the casting design problem in Case-Based Reasoning (CBR). This research investigates a new approach for improving the quality of casting design advice achieved from a CBR system using casting design knowledge associated with past cases. The new approach uses enhanced similarity metrics to those used in previous research in this area to achieve improvements in the advice given. The new similarity metrics proposed here are based on the decomposition of casting shape cases into a set of components. The research into metrics defines and uses the Component Type Similarity Metric (CTM) and Maximum Common Subgraph (MCS) metric between graph representations of the case shapes and are focused on the definition of partial similarity between the components of the same type that take into account the geometrical features and proportions of each single shape component. Additionally, the investigation extends the scope of the research to 3D shapes by defining and evaluating a new metric for the overall similarity between 3D shapes. Additionally, this research investigates a methodology for the integration of the CBR cycle and automation of the feature extraction from target and source case shapes. The ShapeCBR system has been developed to demonstrate the feasibility of integrating the CBR approach for retrieving and reusing casting design advice. The ShapeCBR system automates the decomposition process, the classification process and the shape matching process and is used to evaluate the new similarity metrics proposed in this research and the extension of the approach to 3D shapes. Evaluation of the new similarity metrics show that the efficiency of the system is enhanced using the new similarity metrics and that the new approach provides useful casting design information for 3D casting shapes. Additionally, ShapeCBR shows that it is possible to automate the decomposition and classification of components that allow a case shape to be represented in graph form and thus provide the basis for automating the overall CBR cycle. The thesis concludes with new research questions that emerge from this research and an agenda for further work to be pursued in further research in the area.

Un nombre croissant d’industries ressentent le besoin de formaliser leurs processus d’innovation. Dans ce contexte, les outils du domaine de la qualité et les approches d’aide à la créativité provenant du "brain storming" ont déjà montré leurs limites. Afin de répondre à ces besoins, la TRIZ (Acronyme russe pour Théorie de Résolution des Problèmes Inventifs), développée par l’ingénieur russe G. S. Altshuller au milieu du 20ème siècle, propose une méthode systématique de résolution de problèmes inventifs multidomaines. Selon TRIZ, la résolution de problèmes inventifs consiste en la construction du modèle et l’utilisation des sources de connaissance de la TRIZ. Plusieurs modèles et sources de connaissances permettent la résolution de problèmes inventifs de types différents, comme les quarante Principes Inventifs pour l’élimination des contradictions techniques. Toutes ces sources se situent à des niveaux d’abstractions relativement élevés et sont, donc, indépendantes d’un domaine particulier, qui nécessitent des connaissances approfondies des domaines d’ingénierie différents. Afin de faciliter le processus de résolution de problèmes inventifs, un "Système Intelligent de Gestion de Connaissances" est développé dans cette thèse. D’une part, en intégrant les ontologies des bases de connaissance de la TRIZ, le gestionnaire propose aux utilisateurs de sources de connaissance pertinentes pour le modèle qu’ils construisent, et d’autre part, le gestionnaire a la capacité de remplir "automatiquement" les modèles associés aux autres bases de connaissance. Ces travaux de recherche visent à faciliter et automatiser le processus de résolution de problèmes inventifs. Ils sont basés sur le calcul de similarité sémantique et font usage de différentes technologies provenantes de domaine de l’Ingénierie de Connaissances (modélisation et raisonnement basés sur les ontologies, notamment). Tout d’abord, des méthodes de calcul de similarité sémantique sont proposées pour rechercher et définir les liens manquants entre les bases de connaissance de la TRIZ. Ensuite, les sources de connaissance de la TRIZ sont formalisées comme des ontologies afin de pouvoir utiliser des mécanismes d’inférence heuristique pour la recherche de solutions spécifiques. Pour résoudre des problèmes inventifs, les utilisateurs de la TRIZ choisissent dans un premier temps une base de connaissance et obtiennent une solution abstraite. Ensuite, les éléments des autres bases de connaissance similaires aux éléments sélectionnés dans la première base sont proposés sur la base de la similarité sémantique préalablement calculée. A l’aide de ces éléments et des effets physiques heuristiques, d’autres solutions conceptuelles sont obtenues par inférence sur les ontologies. Enfin, un prototype logiciel est développé. Il est basé sur cette similarité sémantique et les ontologies interviennent en support du processus de génération automatique de solutions conceptuelles
An increasing number of industries feel the need to formalize their innovation processes. In this context, quality domain tools show their limits as well as the creativity assistance approaches derived from brainstorming. TRIZ (Theory of Inventive Problem Solving) appears to be a pertinent answer to these needs. Developed in the middle of the 20th century by G. S. Althshuller, this methodology's goal was initially to improve and facilitate the resolution of technological problems. According to TRIZ, the resolution of inventive problems consists of the construction of models and the use of the corresponding knowledge sources. Different models and knowledge sources were established in order to solve different types of inventive problems, such as the forty inventive principles for eliminating the technical contradictions. These knowledge sources with different levels of abstraction are all built independent of the specific application field, and require extensive knowledge about different engineering domains. In order to facilitate the inventive problem solving process, the development of an "intelligent knowledge manager" is explored in this thesis. On the one hand, according to the TRIZ knowledge sources ontologies, the manager offers to the users the relevant knowledge sources associated to the model they are building. On the other hand, the manager has the ability to fill "automatically" the models of the other knowledge sources. These research works aim at facilitating and automating the process of solving inventive problems based on semantic similarity and ontology techniques. At first, the TRIZ knowledge sources are formalized based on ontologies, such that heuristic inference can be executed to search for specific solutions. Then, methods for calculating semantic similarity are explored to search and define the missing links among the TRIZ knowledge sources. In order to solve inventive problems, the TRIZ user firstly chooses a TRIZ knowledge source to work for an abstract solution. Then, the items of other knowledge sources, which are similar with the selected items of the first knowledge source, are obtained based on semantic similarity calculated in advance. With the help of these similar items and the heuristic physical effects, other specific solutions are returned through ontology inference. Finally, a software prototype is developed based on semantic similarity and ontology inference to support this automatic process of solving inventive problems

Nos travaux de thèses s’intéressent au pronostic de défaillance de composant critique et à l’estimation de la durée de vie résiduelle avant défaillance (RUL). Nous avons développé des méthodes basées sur l’expérience. Cette orientation nous permet de nous affranchir de la définition d’un seuil de défaillance, point problématique lors de l’estimation du RUL. Nous avons pris appui sur le paradigme de Raisonnement à Partir de Cas (R à PC) pour assurer le suivi d’un nouveau composant critique et prédire son RUL. Une approche basée sur les instances (IBL) a été développée en proposant plusieurs formalisations de l’expérience : une supervisée tenant compte de l’ état du composant sous forme d’indicateur de santé et une non-supervisée agrégeant les données capteurs en une série temporelle mono-dimensionnelle formant une trajectoire de dégradation. Nous avons ensuite fait évoluer cette approche en intégrant de la connaissance à ces instances. La connaissance est extraite à partir de données capteurs et est de deux types : temporelle qui complète la modélisation des instances et fréquentielle qui, associée à la mesure de similarité permet d’affiner la phase de remémoration. Cette dernière prend appui sur deux types de mesures : une pondérée entre fenêtres parallèles et fixes et une pondérée avec projection temporelle. Les fenêtres sont glissantes ce qui permet d’identifier et de localiser l’état actuel de la dégradation de nouveaux composants. Une autre approche orientée donnée a été test ée. Celle-ci est se base sur des caractéristiques extraites des expériences, qui sont mono-dimensionnelles dans le premier cas et multi-dimensionnelles autrement. Ces caractéristiques seront modélisées par un algorithme de régression à vecteurs de support (SVR). Ces approches ont été évaluées sur deux types de composants : les turboréacteurs et les batteries «Li-ion». Les résultats obtenus sont intéressants mais dépendent du type de données traitées
Our thesis work is concerned with the development of experience based approachesfor criticalcomponent prognostics and Remaining Useful Life (RUL) estimation. This choice allows us to avoidthe problematic issue of setting a failure threshold.Our work was based on Case Based Reasoning (CBR) to track the health status of a new componentand predict its RUL. An Instance Based Learning (IBL) approach was first developed offering twoexperience formalizations. The first is a supervised method that takes into account the status of thecomponent and produces health indicators. The second is an unsupervised method that fuses thesensory data into degradation trajectories.The approach was then evolved by integrating knowledge. Knowledge is extracted from the sensorydata and is of two types: temporal that completes the modeling of instances and frequential that,along with the similarity measure refine the retrieval phase. The latter is based on two similaritymeasures: a weighted one between fixed parallel windows and a weighted similarity with temporalprojection through sliding windows which allow actual health status identification.Another data-driven technique was tested. This one is developed from features extracted from theexperiences that can be either mono or multi-dimensional. These features are modeled by a SupportVector Regression (SVR) algorithm. The developed approaches were assessed on two types ofcritical components: turbofans and ”Li-ion” batteries. The obtained results are interesting but theydepend on the type of the treated data

Design fixation is a phenomenon experienced by professional designers and engineering design students that stifles creativity and innovation through discouraging ideational productivity, fluency and diversity. During the design idea and concept generation phase of the design process, a reliance on perceptual surface feature similarities between design artefacts increases the likelihood of design fixation leading to design duplication. Psychologists, educators and designers have become increasingly interested in creative idea generation processes that encourage innovation and entrepreneurial outcomes. However, there is a notable lack of collaborative research between psychology, education and engineering design particularly on inductive reasoning of undergraduate engineering students in higher education. The data gathered and analysed for this study provides an insight into property inference decision-making preferences and decision switching (SWITCH) patterns of engineering undergraduates under similarity-based inductive judgements [SIM] and category-based inductive judgements [CAT]. For this psychology experiment, property induction tasks were devised using abstract shapes in a triad configuration. Participants (N = 180), on an undergraduate engineering programme in London, observed a triad of shapes with a target shape more similar-looking to one of two given shapes. Factors manipulated for this experiment included category alignment, category group, property type and target shape. Despite the cognitive development and maturation stage of undergraduate engineers (adults) in higher education, this study identified similarity-based inductive judgements [SIM] to play a significant role during inductive reasoning relative to the strength of category-based inductive judgements [CAT]. In addition to revealing the property inference decision-making preferences of a sample of undergraduate engineers (N = 180), two types of switch classification and two types of non-switch classification (SWITCH) were found and named SIM_NCC, SIM-Salient, Reverse_CAT and CAT_Switching. These different classifications for property inference switching and non-switching presented a more complex pattern of decision-making driven by the relative strength between similarity-based inductive judgements [SIM] and category-based inductive judgements [CAT]. The conditions that encouraged CAT_Switching is of particular interest to design because it corresponds to inference decision switching that affirms the sharing of properties between dissimilar-looking shapes designated as category members, i.e., in a conflicting category alignment condition (CoC). For CAT_Switching, this study found a significant interaction between a particular set of conditions that significantly increased the likelihood of property inference decisions switching to affirm the sharing of properties between dissimilar-looking shapes. Stimuli conditions that combined a conflicting category alignment condition (where dissimilar-looking shapes belong to the same category) with category specificity, a causal property and a target shape with merged (or blended) perceptual surface features significantly increased the likelihood of a property inference decision switching. CAT_Switching has important implications for greater ideational productivity, fluency and diversity to discourage design fixation within the conceptual design space. CAT_Switching conditions could encourage more creative design transformations with alternative design functions through inductive inferences that generalise between dissimilar artefact designs. The findings from this study led to proposing a Cartesian view of the concept design space to represent the possibilities for greater movements through flexible and expanding category boundaries to encourage conceptual combinations, greater ideational fluency and greater ideational diversity within a configuration design space. This study has also created a platform for further research into property inference decision-making, ideational diversity and category boundary flexibility under stimuli conditions that encourage designers and design students to make inductive generalisations between dissimilar domains of knowledge through a greater emphasis on causal relations and semantic networks.

碩士
國立交通大學
資訊科學系
88
Case-based reasoning (CBR) is a methodology of problem-solving in artificial intelligence. Just like human being, CBR uses prior cases to find out suitable solution for the new problems. Unlike the others, CBR pays attention to the characteristics of each case. CBR can correctly take advantage of the situations and methods in former cases to solve problems. A critical task of CBR is to retrieve similar prior cases accurately and many researchers have proposed some useful technologies to handle such problem. However, increasingly larger number of cases influences the performance of retrieving similar cases for the large-scale CBR was seldom been discussed. In this thesis, the performance issue of large-scale CBR is discussed and a new indexing method, called bit-wise indexing method, and the corresponding efficient algorithms are proposed for retrieving the similar cases in large-scale CBR efficiently. The bit-wise indexing method and the corresponding algorithm can be easily parallelized and thus gets great performance improvement in case retrieving and similarity measuring. Some experiments are made for comparing the performance with other methods and the results show the performance of proposed method is admirable.

A Stroke can affect different parts of the human body depending on the area of brain effected; our research focuses on upper limb motor dysfunction for stroke patients. In current practice, ordinal scale systems are used for conducting physical assessment of upper limb impairment. The reliability of these assessments is questionable, since their coarse ratings cannot reliably distinguish between the different levels of performance. This thesis describes the design, implementation and evaluation of a novel system to facilitate stroke diagnosis which relies on data collected with an innovative KINARM robotic tool. This robotic tool allows for an objective quantification of motor function and performance assessment for stroke patients. The main methodology for the research is Case Based Reasoning (CBR) - an effective paradigm of artificial intelligence that relies on the principle that a new problem is solved by observing similar, previously encountered problems and adapting their known solutions. A CBR system was designed and implemented for a repository of stroke subjects who had an explicit diagnosis and prognosis. For a new stroke patient, whose diagnosis was yet to be confirmed and who had an indefinite prognosis, the CBR model was effectively used to retrieve analogous cases of previous stroke patients. These similar cases provide useful information to the clinicians, facilitating them in reaching a potential solution for stroke diagnosis and also a means to validate other imaging tests and clinical assessments to confirm the diagnosis and prognosis.
Thesis (Master, Computing) -- Queen's University, 2008-09-27 11:14:04.85

Dissertação de mestrado em Engenharia Informática
On a business context, it is responsibility of the Software Product Support Team analyze and solve, if necessary, problems that may arise on software products. Sometimes, the reported problems are not a real defect, i.e., sometimes the client does not have a full understanding about all features of the software product. The team must evaluate and analyze all the Problem Reports that arrive every day. As products are spread across different customers, it is normal to have Problem Reports that are very similar to others that have already been solved for other clients and/or by another member of the Support Team. This dissertation proposes the development of a system that is able to analyze a Problem Report and then provide past problems that are similar to the one being analyzed. An artificial intelligence technique, named Case-Based Reasoning, will be used to achieve such goals. Existent Case-Based Reasoning systems are neither complete nor adaptable to specific domains since the effort to adapt either the reasoning process or the knowledge representation mechanism, to a new domain, is too high. To address such drawbacks, a generic reasoning component will be designed and developed. This dissertation introduces a new approach to the typical Case-Based Reasoning cycle where is possible to handle default, unknown and incomplete data.
Num contexto empresarial, é da responsabilidade da equipa de Suporte ao Produto de Software analisar e corrigir, se necessário, problemas que possam surgir em produtos de software. Muitas vezes esses mesmos problemas não o são, isto é, muitas vezes é o cliente que não tem uma percepção completa do funcionamento do produto. A equipa deve avaliar e analisar todos os Problem Reports que vão chegando dia após dia. Como os produtos se encontram espalhados por diferentes clientes, é normal aparecerem problemas que são muito semelhantes com outros que já foram resolvidos para outros clientes e/ou por um outro membro da equipa de Suporte. Esta dissertação propõe o desenvolvimento de um sistema que seja capaz de analisar um problema e indicar problemas passados que sejam semelhantes com o que se está analisar. Uma técnica da inteligência artificial, de seu nome Raciocínio Baseado em Casos, será utilizada para atingir tais objectivos. Os sistemas já existentes que usam esta técnica não são completos nem adaptáveis a domínios específicos, uma vez que o esforço para adaptar tanto o processo de raciocínio como a representação do conhecimento, para um novo domínio, é demasiado elevado. Para solucionar tais problemas, um componente de raciocínio genérico será especificado e desenvolvido. Esta dissertação introduz uma nova abordagem ao ciclo do Raciocínio-Baseado em Casos onde é possível tratar informação desconhecida.

The notion of similarity plays an important role in machine learning and artificial intelligence. It is widely used in tasks related to a supervised classification, clustering, an outlier detection and planning. Moreover, in domains such as information retrieval or case-based reasoning, the concept of similarity is essential as it is used at every phase of the reasoning cycle. The similarity itself, however, is a very complex concept that slips out from formal definitions. A similarity of two objects can be different depending on a considered context. In many practical situations it is difficult even to evaluate the quality of similarity assessments without considering the task for which they were performed. Due to this fact the similarity should be learnt from data, specifically for the task at hand. In this dissertation a similarity model, called Rule-Based Similarity, is described and an algorithm for constructing this model from available data is proposed. The model utilizes notions from the rough set theory to derive a similarity function that allows to approximate the similarity relation in a given context. The construction of the model starts from the extraction of sets of higher-level features. Those features can be interpreted as important aspects of the similarity. Having defined such features it is possible to utilize the idea of Tversky’s feature contrast model in order to design an accurate and psychologically plausible similarity function for a given problem. Additionally, the dissertation shows two extensions of Rule-Based Similarity which are designed to efficiently deal with high dimensional data. They incorporate a broader array of similarity aspects into the model. In the first one it is done by constructing many heterogeneous sets of features from multiple decision reducts. To ensure their diversity, a randomized reduct computation heuristic is proposed. This approach is particularly well-suited for dealing with the few-objects-many-attributes problem, e.g. the analysis of DNA microarray data. A similar idea can be utilized in the text mining domain. The second of the proposed extensions serves this particular purpose. It uses a combination of a semantic indexing method and an information bireducts computation technique to represent texts by sets of meaningful concepts. The similarity function of the proposed model can be used to perform an accurate classification of previously unseen objects in a case-based fashion or to facilitate clustering of textual documents into semantically homogeneous groups. Experiments, whose results are also presented in the dissertation, show that the proposed models can successfully compete with the state-of-the-art algorithms.
Pojęcie podobieństwa pełni istotną rolę w dziedzinach uczenia maszynowego i sztucznej inteligencji. Jest ono powszechnie wykorzystywane w zadaniach dotyczących nadzorowanej klasyfikacji, grupowania, wykrywania nietypowych obiektów oraz planowania. Ponadto w dziedzinach takich jak wyszukiwanie informacji (ang. information retrieval) lub wnioskowanie na podstawie przykładów (ang. case-based reasoning) pojęcie podobieństwa jest kluczowe ze względu na jego obecność na wszystkich etapach wyciągania wniosków. Jednakże samo podobieństwo jest pojęciem niezwykle złożonym i wymyka się próbom ścisłego zdefiniowania. Stopień podobieństwa między dwoma obiektami może być różny w zależności od kontekstu w jakim się go rozpatruje. W praktyce trudno jest nawet ocenić jakość otrzymanych stopni podobieństwa bez odwołania się do zadania, któremu mają służyć. Z tego właśnie powodu modele oceniające podobieństwo powinny być wyuczane na podstawie danych, specjalnie na potrzeby realizacji konkretnego zadania. W niniejszej rozprawie opisano model podobieństwa zwany Regułowym Modelem Podobieństwa (ang. Rule-Based Similarity) oraz zaproponowano algorytm tworzenia tego modelu na podstawie danych. Wykorzystuje on elementy teorii zbiorów przybliżonych do konstruowania funkcji podobieństwa pozwalającej aproksymować podobieństwo w zadanym kontekście. Konstrukcja ta rozpoczyna się od wykrywania zbiorów wysokopoziomowych cech obiektów. Mogą być one interpretowane jako istotne aspekty podobieństwa. Mając zdefiniowane tego typu cechy możliwe jest wykorzystanie idei modelu kontrastu cech Tversky’ego (ang. feature contrast model) do budowy precyzyjnej oraz zgodnej z obserwacjami psychologów funkcji podobieństwa dla rozważanego problemu. Dodatkowo, niniejsza rozprawa zawiera opis dwóch rozszerzeń Regułowego Modelu Podobieństwa przystosowanych do działania na danych o bardzo wielu atrybutach. Starają się one włączyć do modelu szerszy zakres aspektów podobieństwa. W pierwszym z nich odbywa się to poprzez konstruowanie wielu zbiorów cech z reduktów decyzyjnych. Aby zapewnić ich zróżnicowanie, zaproponowano algorytm łączący heurystykę zachłanna z elementami losowymi. Podejście to jest szczególnie wskazane dla zadań związanych z problemem małej liczby obiektów i dużej liczby cech (ang. the few-objects-many-attributes problem), np. analizy danych mikromacierzowych. Podobny pomysł może być również wykorzystany w dziedzinie analizy tekstów. Realizowany jest on przez drugie z proponowanych rozszerzeń modelu. Łączy ono metodę semantycznego indeksowania z algorytmem obliczania bireduktów informacyjnych, aby reprezentować teksty dobrze zdefiniowanymi pojęciami. Funkcja podobieństwa zaproponowanego modelu może być wykorzystana do klasyfikacji nowych obiektów oraz do łączenia dokumentów tekstowych w semantycznie spójne grupy. Eksperymenty, których wyniki opisano w rozprawie, dowodzą, ze zaproponowane modele mogą skutecznie konkurować nawet z powszechnie uznanymi rozwiązaniami.

碩士
國立成功大學
工業與資訊管理學系專班
101
Supplier selection is one of the most important jobs of supply chain management. Because of the reducing of product life cycle, manufacturers have to utilize the existing supply chain to shorten the time from design to market as well as improve their competition. In fact, buyers will nominate the potential suppliers by their expertise and experience to decide the most appropriate candidates based on the RFQ (request for quotation) and negotiation result. However, different buyers lead to different potential supplier list and supplier evaluation outcomes. Therefore, it becomes an important issue to find the most appropriate supplier regardless of the differences of the expertise and experience of buyers. Similar to the purchasing process, we propose a two-stage method for supplier selection. The first stage employs the case-based reasoning method to find potential suppliers by calculating the cosine similarity of new material description with the old ones stored in a database. Then the technique for order preference by similarity to ideal solution (TOPSIS) is used in the second stage to rank the potential suppliers by their performance that is evaluated by the following factors: quality, delivery, cost, and supply possibility. The weights of the factors are set up in advance by decision makers. For the case analyzed in this study, the most appropriate supplier determined by the TOPSIS method is identical to the one suggested by a senior professional buyer. This demonstrates that the method proposed by this study can be helpful to the buyers for supplier selection.

碩士
國立臺灣科技大學
資訊工程系
103
Fuzzy interpolative reasoning is a very important research topic for sparse fuzzy rule-based systems. In this thesis, we propose two new fuzzy interpolative reasoning methods for sparse fuzzy rule-based systems based on polygonal fuzzy sets and the ranking values of polygonal fuzzy sets. In the first method of our thesis, we propose a new fuzzy interpolative reasoning method for sparse fuzzy rule-based systems based on ranking values of polygonal fuzzy sets and automatically generated weights of fuzzy rules. The experimental results show that the proposed method can overcome the drawbacks of the existing fuzzy interpolative reasoning methods for fuzzy interpolative reasoning in sparse fuzzy rule-based systems. In the second method of our thesis, we propose a new adaptive fuzzy interpolation method based on ranking values of polygonal fuzzy sets and similarity measures between polygonal fuzzy sets. The proposed adaptive fuzzy interpolation method performs fuzzy interpolative reasoning using multiple fuzzy rules with multiple antecedent variables and solves the contradictions after the fuzzy interpolative reasoning processes based on similarity measures between polygonal fuzzy sets. The experimental results show that the proposed adaptive fuzzy interpolation method outperforms the existing methods for fuzzy interpolative reasoning in sparse fuzzy rule-based systems.