Dissertations / Theses on the topic 'Large Scale Recommendation'

The Internet and the World Wide Web have taken over as the standard reading and finding news. This makes it possible for news readers to carefully choose the news that is most interesting for them. Due to the large amounts of articles, it can be a challenging and time consuming task to find the wanted information. Simplifying this process for the news readers would be beneficial.This thesis explores the idea of filtering out unwanted news articles and serving the useful ones to the reader through mobile platforms. It is part of a bigger project named SmartMedia that focuses on using complex strategies for delivering news to the users. While the overall strategy is based on using the total context of users to serve new, the specific scope of this thesis is creating user profiles from user acts logged by the system. The motivation is to utilize these profiles in cooperation with information filtering techniques to help reach the overall goal.A big part of this thesis focuses on implementing Hadoop jobs that summarizes the user logs into profiles. In the solution, each user profile consists of two vectors. A category vector that describes the user?s interests in the different news categories and a keyword vector that exploits entities defined in news articles to analyse at a low granularity level. The results are evaluated and discussed at the end.How to evaluate the effectiveness and accuracy of the user profiles is difficult. Little real data was available during this research and actual data is needed. Data that replicates real users is hard to forge and is needed for both evaluation and calibration of the implementation. Thus, the focus of the discussion is on how to perform these two tasks when the system is deployed.

Cette thèse s'intéresse au problème de l'apprentissage à partir d'interactions en utilisant le cadre du bandit contextuel hors ligne. En particulier, nous nous intéressons à deux sujets connexes : (1) l'apprentissage de politiques hors ligne avec des certificats de performance, et (2) l'apprentissage rapide et efficace de politiques, pour le problème de recommandation à grande échelle. Pour (1), nous tirons d'abord parti des résultats du cadre d'optimisation distributionnellement robuste pour construire des bornes asymptotiques, sensibles à la variance, qui permettent l'évaluation des performances des politiques. Ces bornes nous aident à obtenir de nouveaux objectifs d'apprentissage plus pratiques grâce à leur nature composite et à leur calibrage simple. Nous analysons ensuite le problème d'un point de vue PAC-Bayésien et fournissons des bornes, plus étroites, sur les performances des politiques. Nos résultats motivent de nouvelles stratégies, qui offrent des certificats de performance sur nos politiques avant de les déployer en ligne. Les stratégies nouvellement dérivées s'appuient sur des objectifs d'apprentissage composites qui ne nécessitent pas de réglage supplémentaire. Pour (2), nous proposons d'abord un modèle bayésien hiérarchique, qui combine différents signaux, pour estimer efficacement la qualité de la recommandation. Nous fournissons les outils computationnels appropriés pour adapter l'inférence aux problèmes à grande échelle et démontrons empiriquement les avantages de l'approche dans plusieurs scénarios. Nous abordons ensuite la question de l'accélération des approches communes d'optimisation des politiques, en nous concentrant particulièrement sur les problèmes de recommandation avec des catalogues de millions de produits. Nous dérivons des méthodes d'optimisation, basées sur de nouvelles approximations du gradient calculées en temps logarithmique par rapport à la taille du catalogue. Notre approche améliore le temps linéaire des méthodes courantes de calcul de gradient, et permet un apprentissage rapide sans nuire à la qualité des politiques obtenues
This thesis presents contributions to the problem of learning from logged interactions using the offline contextual bandit framework. We are interested in two related topics: (1) offline policy learning with performance certificates, and (2) fast and efficient policy learning applied to large scale, real world recommendation. For (1), we first leverage results from the distributionally robust optimisation framework to construct asymptotic, variance-sensitive bounds to evaluate policies' performances. These bounds lead to new, more practical learning objectives thanks to their composite nature and straightforward calibration. We then analyse the problem from the PAC-Bayesian perspective, and provide tighter, non-asymptotic bounds on the performance of policies. Our results motivate new strategies, that offer performance certificates before deploying the policies online. The newly derived strategies rely on composite learning objectives that do not require additional tuning. For (2), we first propose a hierarchical Bayesian model, that combines different signals, to efficiently estimate the quality of recommendation. We provide proper computational tools to scale the inference to real world problems, and demonstrate empirically the benefits of the approach in multiple scenarios. We then address the question of accelerating common policy optimisation approaches, particularly focusing on recommendation problems with catalogues of millions of items. We derive optimisation routines, based on new gradient approximations, computed in logarithmic time with respect to the catalogue size. Our approach improves on common, linear time gradient computations, yielding fast optimisation with no loss on the quality of the learned policies

A pressing need for efficient personalized recommendations has emerged in crowdsourcing systems. On the one hand, workers confront a flood of tasks, and they often spend too much time to find tasks matching their skills and interests. Thus, workers want effective recommendation of the most suitable tasks with regard to their skills and preferences. On the other hand, requesters sometimes receive results in low-quality completion since a less qualified worker may start working on a task before a better-skilled worker may get hands on. Thus, requesters want reliable recommendation of the best workers for their tasks in terms of workers' qualifications and accountability. The task and worker recommendation problems in crowdsourcing systems have brought up unique characteristics that are not present in traditional recommendation scenarios, i.e., the huge flow of tasks with short lifespans, the importance of workers' capabilities, and the quality of the completed tasks. These unique features make traditional recommendation approaches (mostly developed for e-commerce markets) no longer satisfactory for task and worker recommendation in crowdsourcing systems. In this research, we reveal our insight into the essential difference between the tasks in crowdsourcing systems and the products/items in e-commerce markets, and the difference between buyers' interests in products/items and workers' interests in tasks. Our insight inspires us to bring up categories as a key mediation mechanism between workers and tasks. We propose a two-tier data representation scheme (defining a worker-category suitability score and a worker-task attractiveness score) to support personalized task and worker recommendation. We also extend two optimization methods, namely least mean square error (LMS) and Bayesian personalized rank (BPR) in order to better fit the characteristics of task/worker recommendation in crowdsourcing systems. We then integrate the proposed representation scheme and the extended optimization methods along with the two adapted popular learning models, i.e., matrix factorization and kNN, and result in two lines of top-N recommendation algorithms for crowdsourcing systems: (1) Top-N-Tasks (TNT) recommendation algorithms for discovering the top-N most suitable tasks for a given worker, and (2) Top-N-Workers (TNW) recommendation algorithms for identifying the top-N best workers for a task requester. An extensive experimental study is conducted that validates the effectiveness and efficiency of a broad spectrum of algorithms, accompanied by our analysis and the insights gained.

La dynamique humaine est un sujet essentiel de l'informatique centrée sur l’homme. Elle se concentre sur la compréhension des régularités sous-jacentes, des relations, et des changements dans les comportements humains. En analysant la dynamique humaine, nous pouvons comprendre non seulement des comportements individuels, tels que la présence d’une personne à un endroit précis, mais aussi des comportements collectifs, comme les mouvements sociaux. L’exploration de la dynamique humaine permet ainsi diverses applications, entre autres celles des services géo-dépendants personnalisés dans des scénarios de ville intelligente. Avec l'omniprésence des smartphones équipés de GPS, les réseaux sociaux de géolocalisation ont acquis une popularité croissante au cours des dernières années, ce qui rend les données de comportements des utilisateurs disponibles à grande échelle. Sur les dits réseaux sociaux de géolocalisation, les utilisateurs peuvent partager leurs activités en temps réel avec par l'enregistrement de leur présence à des points d'intérêt (POIs), tels qu’un restaurant. Ces données d'activité contiennent des informations massives sur la dynamique humaine. Dans cette thèse, nous explorons la dynamique humaine basée sur les données massives des réseaux sociaux de géolocalisation. Concrètement, du point de vue individuel, nous étudions la préférence de l'utilisateur quant aux POIs avec des granularités différentes et ses applications, ainsi que la régularité spatio-temporelle des activités des utilisateurs. Du point de vue collectif, nous explorons la forme d'activité collective avec les granularités de pays et ville, ainsi qu’en corrélation avec les cultures globales
Human dynamics is an essential aspect of human centric computing. As a transdisciplinary research field, it focuses on understanding the underlying patterns, relationships, and changes of human behavior. By exploring human dynamics, we can understand not only individual’s behavior, such as a presence at a specific place, but also collective behaviors, such as social movement. Understanding human dynamics can thus enable various applications, such as personalized location based services. However, before the availability of ubiquitous smart devices (e.g., smartphones), it is practically hard to collect large-scale human behavior data. With the ubiquity of GPS-equipped smart phones, location based social media has gained increasing popularity in recent years, making large-scale user activity data become attainable. Via location based social media, users can share their activities as real-time presences at Points of Interests (POIs), such as a restaurant or a bar, within their social circles. Such data brings an unprecedented opportunity to study human dynamics. In this dissertation, based on large-scale location centric social media data, we study human dynamics from both individual and collective perspectives. From individual perspective, we study user preference on POIs with different granularities and its applications in personalized location based services, as well as the spatial-temporal regularity of user activities. From collective perspective, we explore the global scale collective activity patterns with both country and city granularities, and also identify their correlations with diverse human cultures

La dynamique humaine est un sujet essentiel de l'informatique centrée sur l’homme. Elle se concentre sur la compréhension des régularités sous-jacentes, des relations, et des changements dans les comportements humains. En analysant la dynamique humaine, nous pouvons comprendre non seulement des comportements individuels, tels que la présence d’une personne à un endroit précis, mais aussi des comportements collectifs, comme les mouvements sociaux. L’exploration de la dynamique humaine permet ainsi diverses applications, entre autres celles des services géo-dépendants personnalisés dans des scénarios de ville intelligente. Avec l'omniprésence des smartphones équipés de GPS, les réseaux sociaux de géolocalisation ont acquis une popularité croissante au cours des dernières années, ce qui rend les données de comportements des utilisateurs disponibles à grande échelle. Sur les dits réseaux sociaux de géolocalisation, les utilisateurs peuvent partager leurs activités en temps réel avec par l'enregistrement de leur présence à des points d'intérêt (POIs), tels qu’un restaurant. Ces données d'activité contiennent des informations massives sur la dynamique humaine. Dans cette thèse, nous explorons la dynamique humaine basée sur les données massives des réseaux sociaux de géolocalisation. Concrètement, du point de vue individuel, nous étudions la préférence de l'utilisateur quant aux POIs avec des granularités différentes et ses applications, ainsi que la régularité spatio-temporelle des activités des utilisateurs. Du point de vue collectif, nous explorons la forme d'activité collective avec les granularités de pays et ville, ainsi qu’en corrélation avec les cultures globales
Human dynamics is an essential aspect of human centric computing. As a transdisciplinary research field, it focuses on understanding the underlying patterns, relationships, and changes of human behavior. By exploring human dynamics, we can understand not only individual’s behavior, such as a presence at a specific place, but also collective behaviors, such as social movement. Understanding human dynamics can thus enable various applications, such as personalized location based services. However, before the availability of ubiquitous smart devices (e.g., smartphones), it is practically hard to collect large-scale human behavior data. With the ubiquity of GPS-equipped smart phones, location based social media has gained increasing popularity in recent years, making large-scale user activity data become attainable. Via location based social media, users can share their activities as real-time presences at Points of Interests (POIs), such as a restaurant or a bar, within their social circles. Such data brings an unprecedented opportunity to study human dynamics. In this dissertation, based on large-scale location centric social media data, we study human dynamics from both individual and collective perspectives. From individual perspective, we study user preference on POIs with different granularities and its applications in personalized location based services, as well as the spatial-temporal regularity of user activities. From collective perspective, we explore the global scale collective activity patterns with both country and city granularities, and also identify their correlations with diverse human cultures

To support the continuously increasing number of mobile telephone users around the world, mobile communication systems have become more advanced and sophisticated in their designs. As a result of the great success with the second generation mobile radio networks, deployment of the third and development of fourth generations, the demand for higher data rates to support available services, such as internet connection, video telephony and personal navigation systems, is ever growing. To be able to meet the requirements regarding bandwidth and number of users, enhancements of existing systems and introductions of conceptually new technologies and techniques have been researched and developed. Although new proposed technologies in theory provide increased network capacity, the backbone of a successful roll-out of a mobile telephone system is inevitably the planning of the network’s cellular structure. Hence, the fundamental aspect to a reliable cellular planning is the knowledge about the physical radio channel for wide sets of different propagation scenarios. Therefore, to study radio wave propagation in typical Australian environments, the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and the Australian Telecommunications Cooperative Research Centre (ATcrc) in collaboration developed a cellular code division multiple access (CDMA) pilot scanner. The pilot scanner measurement equipment enables for radio wave propagation measurements in available commercial CDMA mobile radio networks, which in Australia are usually deployed for extensive rural areas. Over time, the collected measurement data has been used to characterise many different types of mobile radio environments and some of the results are presented in this thesis. The thesis is divided into an introduction section and four parts based on peer-reviewed international research publications. The introduction section presents the reader with some relevant background on channel and propagation modelling. Also, the CDMA scanner measurement system that was developed in parallel with the research results founding this thesis is presented. The first part presents work on the evaluation and development of the different revisions of the Recommendation ITU-R P.1546 point-to-area radio wave propagation prediction model. In particular, the modified application of the terrain clearance angle (TCA) and the calculation method of the effective antenna height are scrutinized. In the second part, the correlation between the smallscale fading characteristics, described by the Ricean K-factor, and the vegetation density in the vicinity of the mobile receiving antenna is investigated. The third part presents an artificial neural network (ANN) based technique incorporated to predict path loss in rural macrocell environments. Obtained results, such as prediction accuracy and training time, are presented for different sized ANNs and different training approaches. Finally, the fourth part proposes an extension of the path loss ANN enabling the model to also predict small-scale fading characteristics.

In this thesis, we describe our design of Feature Recommender, a Mozilla Firefox browser extension, which proactively recommends features that it predicts will benefit users based on their individual usage behaviors. The goal of these pop-up notifications is to help users discover new features. How to maximize the effectiveness of such notifications while minimizing user interruptions remains a difficult open problem. One approach is to carefully time when the notifications are delivered. In our deployment of Feature Recommender, we study the effect of two delivery timing parameters: delivery rate and the user's context at the moment of delivery. We also investigate the effect of prediction algorithm sensitivity. We conducted three field studies, each about 4 weeks: (1) A preliminary study (N=10) to determine reasonable interruptible-moments; (2) A qualitative study (N=20) to assess the design and effectiveness of our extension; and (3) A near-identical study (N= ~3K) to assess quantitatively the effect of the timing parameters. Across all conditions Feature Recommender helped users adopt on average 18% of the features they were recommended, and as many as 24% when they were delivered at random times with a 1-per-day delivery rate limit. We show that lack of trust in recommendations is a key factor in hindering their effectiveness.
Science, Faculty of
Computer Science, Department of
Graduate

MICROFICHE COPY AVAILABLE IN ARCHIVES AND ROTCH.
Thesis (M. Arch.)--Massachusetts Institute of Technology, Dept. of Architecture; and, (M.C.P.)--Massachusetts Institute of Technology, Dept. of Urban Studies and Planning, 1981.
Bibliography: leaves 173-176.
by James Rutherford Richardson, IV.

Made available in DSpace on 2016-04-28T20:56:30Z (GMT). No. of bitstreams: 1 Antonio Vanderlei Tavares.pdf: 584517 bytes, checksum: 338414fd05f955f86db1005d6ab0bd33 (MD5) Previous issue date: 2012-06-14
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior
This study aimed to identify and analyze recommendations and decisions from the large-scale assessment of a system of private education in the state of Sao Paulo. The methodological trajectory, initially, there were readings of the final reports of large-scale assessments to identify and document the recommendations and decisions taken in the period 1999 to 2008. Then, set up categories for the analysis of recommendations and decisions. These categories were based on the assumptions that conceive of educational assessment as a tool for monitoring and improving the quality of education offered in school systems. These assumptions were based on the principles of democratic and participative management. The analysis of the recommendations of evaluations allowed us to observe that there is a group of recommendations on teachers' work and one that is intended to administration. However, both contain recommendations that call for analysis and planning by the management team of the education system. In respect of decisions shows that there was a greater concern with the development of actions to curriculum and continuing education of educators. No actions have been identified related to contextual variables raised by the assessment scale. Both the analysis of the recommendations described in assessment reports, when the decisions indicated that the evaluation of large scale offers significant contributions on the education system, but they require planning actions intermediate between the schools and central management
Este trabalho teve o objetivo de identificar e analisar recomendações e decisões tomadas a partir da avaliação de larga escala de um sistema de ensino particular do estado de São Paulo. Na trajetória metodológica, inicialmente, foram realizadas leituras dos relatórios finais das avaliações de larga escala e de documentos para identificarmos as recomendações e as decisões tomadas no período de 1999 a 2008. Em seguida, definiram-se categorias para a análise das recomendações e decisões tomadas. Essas categorias foram elaboradas com base nos pressupostos que concebem a avaliação educacional como um instrumento para monitoramento e aprimoramento da qualidade da educação oferecida nos sistemas de ensino. Tais pressupostos foram fundamentados nos princípios da gestão democrática e participativa. As análises das recomendações das avaliações nos permitiram observar que há um grupo de recomendações sobre o trabalho docente e outro que se destina à gestão. No entanto, ambos contêm recomendações que suscitam análise e planejamento por parte da equipe gestora do sistema educativo. Quanto às decisões tomadas, observamos que houve uma preocupação maior com o desenvolvimento de ações voltadas ao currículo e à formação continuada dos educadores. Não foram identificadas ações relacionadas com variáveis contextuais levantadas pela avaliação de larga escala. Tanto as análises das recomendações descritas nos relatórios das avaliações, quando das decisões tomadas indicaram que a avaliação de larga escala oferece contribuições relevantes sobre o sistema de ensino, porém estas necessitam de ações de planejamento intermediárias entre as escolas e a gestão central

Les systèmes de recommandation (RS) et le pair-à-pair (P2) sont complémen-taires pour faciliter le partage de données à grande échelle: RS pour filtrer et person-naliser les requêtes des utilisateurs, et P2P pour construire des systèmes de partage de données décentralisés à grande échelle. Cependant, il reste beaucoup de difficultés pour construire des RS efficaces dans une infrastructure P2P. Dans cette thèse, nous considérons des communautés en ligne à grande échelle, où les utilisateurs notent les contenus qu'ils explorent et gardent dans leur espace de travail local les contenus de qualité pour leurs sujets d'intérêt. Notre objectif est de construire un P2P-RS efficace pour ce contexte. Nous exploitons les sujets d'intérêt des utilisateurs (extraits automatiquement des contenus et de leurs notes) et les don-nées sociales (amitié et confiance) afin de construire et maintenir un overlay P2P so-cial. La thèse traite de plusieurs problèmes. D'abord, nous nous concentrons sur la conception d'un P2P-RS qui passe à l'échelle, appelé P2Prec, en combinant les ap-proches de recommandation par filtrage collaboratif et par filtrage basé sur le contenu. Nous proposons alors de construire et maintenir un overlay P2P dynamique grâce à des protocoles de gossip. Nos résultats d'expérimentation montrent que P2Prec per-met d'obtenir un bon rappel avec une charge de requêtes et un trafic réseau accep-tables. Ensuite, nous considérons une infrastructure plus complexe afin de construire et maintenir un overlay P2P social, appelé F2Frec, qui exploite les relations sociales entre utilisateurs. Dans cette infrastructure, nous combinons les aspects filtrage par contenu et filtrage basé social, pour obtenir un P2P-RS qui fournit des résultats de qualité et fiables. A l'aide d'une évaluation de performances extensive, nous mon-trons que F2Frec améliore bien le rappel, ainsi que la confiance dans les résultats avec une surcharge acceptable. Enfin, nous décrivons notre prototype de P2P-RS que nous avons implémenté pour valider notre proposition basée sur P2Prec et F2Frec.

碩士
國立暨南國際大學
資訊工程學系
96
Recommendation systems based on collaborative filtering predict customer preference for items by learning the customer-item pairs of the past. A predominant approach to collaborative filtering is neighborhood based (KNN-based), where the customer-item preference rating is discovered from ratings of similar items or customers. In this research work, we apply latent semantic analysis of IR to discover relation densities between customers and items. Unlike previous approaches, this method does not require system to perform exhaustive search for association rules nor need to build a predefine knowledge base. This analysis is done by pure mathematic method. During the math procedures, users will be able to decide the size of data to be analyzed depending on their actual circumstances. The preliminary experiment results show an encouraging accuracy over 90%. To further deal with large-scale, sparse data, we apply SVM and feature selection techniques to reduce the dimension of data representation. SVM allows categorizing data samples into several subsets, in which the data samples are smaller. We apply feature selection in those smaller data samples to filter out less important or less relevant information that help fit the RAM size on ordinary PCs.

Thesis (M.S.)--University of Wisconsin--Madison, 2002.
Typescript. eContent provider-neutral record in process. Description based on print version record. Includes bibliographical references (p. 159-171).