With more and more images being uploaded to social networks each day, the resources for identifying a large portion of the world are available. However the tools to harness and utilize this information are not sufficient. This thesis presents a system, called PhacePhinder, which can build a face database from a social network and have it accessible from mobile devices. Through combining existing technologies, this is made possible. It also makes use of a fusion probabilistic latent semantic analysis to determine strong connections between users and content. Using this information we can determine the most meaningful social connection to a recognized person, allowing us to inform the user of how they know the person being recognized. We conduct a series of offline and user tests to verify our results and compare them to existing algorithms. We show, that through combining a user’s friendship information as well as picture occurrence information, we can make stronger recommendations than based on friendship alone. We demonstrate a working prototype that can identify a face from a picture taken from a mobile phone, using a database derived from images gathered directly from a social network, and return a meaningful social connection to the recognized face.

With more and more images being uploaded to social networks each day, the resources for identifying a large portion of the world are available. However the tools to harness and utilize this information are not sufficient. This thesis presents a system, called PhacePhinder, which can build a face database from a social network and have it accessible from mobile devices. Through combining existing technologies, this is made possible. It also makes use of a fusion probabilistic latent semantic analysis to determine strong connections between users and content. Using this information we can determine the most meaningful social connection to a recognized person, allowing us to inform the user of how they know the person being recognized. We conduct a series of offline and user tests to verify our results and compare them to existing algorithms. We show, that through combining a user’s friendship information as well as picture occurrence information, we can make stronger recommendations than based on friendship alone. We demonstrate a working prototype that can identify a face from a picture taken from a mobile phone, using a database derived from images gathered directly from a social network, and return a meaningful social connection to the recognized face.

With more and more images being uploaded to social networks each day, the resources for identifying a large portion of the world are available. However the tools to harness and utilize this information are not sufficient. This thesis presents a system, called PhacePhinder, which can build a face database from a social network and have it accessible from mobile devices. Through combining existing technologies, this is made possible. It also makes use of a fusion probabilistic latent semantic analysis to determine strong connections between users and content. Using this information we can determine the most meaningful social connection to a recognized person, allowing us to inform the user of how they know the person being recognized. We conduct a series of offline and user tests to verify our results and compare them to existing algorithms. We show, that through combining a user’s friendship information as well as picture occurrence information, we can make stronger recommendations than based on friendship alone. We demonstrate a working prototype that can identify a face from a picture taken from a mobile phone, using a database derived from images gathered directly from a social network, and return a meaningful social connection to the recognized face.

Este documento apresenta um estudo que avalia o uso de informação temporal na tarefa de recomendação de tweets no twitter. Foram explorados dois aspectos temporais: a vida útil de tópico de informação e a sua versão personalizada para cada usuário. A aplicação destes aspectos temporais foi avaliada utilizando-se três sistemas de recomendação implementados. Também avaliamos dois modelos de tópicos utilizados para representar tweets: o modelo bag of words e um modelo de tópicos latentes extraídos por LDA (Latent Dirichlet Allocation). Além disso, avaliamos o uso de máquinas de vetor de suporte para estimar o perfil de interesses de usuário, comparando esta abordagem com uma outra mais simples. Os experimentos foram executados utilizando-se um conjunto de dados com 414 milhões de tweets publicados por 321 mil usuários. Os resultados apresentados demonstram que o uso de vida útil de tópico na tarefa de recomendação melhora a qualidade das recomendações, e o uso da versão personalizada desta informação melhorou ainda mais a qualidade destas / This document presents a study that evaluates the use of temporal information in the task of recommending tweets on Twitter. Two temporal aspects have been analysed: the lifespan of information topic and its personalized version for each user. The application of such temporal aspects has been evaluated using three recommendation systems implemented in this work. We also evaluated two topic models considered to describe tweets: a bag of words model and a model of latent topics extracted using LDA (Latent Dirichlet Allocation). Furthermore, we evaluated the use of SVM (Support Vector Machines) to estimate the user profile, comparing this approach with a simpler one. The experiments have been executed using a dataset with 414 millions of tweets published by 321 thousands of users. The results show that the use of topic lifespan information increases the quality of recommendation, and the personalized version of this information increases the quality even more

Este documento apresenta um estudo que avalia o uso de informação temporal na tarefa de recomendação de tweets no twitter. Foram explorados dois aspectos temporais: a vida útil de tópico de informação e a sua versão personalizada para cada usuário. A aplicação destes aspectos temporais foi avaliada utilizando-se três sistemas de recomendação implementados. Também avaliamos dois modelos de tópicos utilizados para representar tweets: o modelo bag of words e um modelo de tópicos latentes extraídos por LDA (Latent Dirichlet Allocation). Além disso, avaliamos o uso de máquinas de vetor de suporte para estimar o perfil de interesses de usuário, comparando esta abordagem com uma outra mais simples. Os experimentos foram executados utilizando-se um conjunto de dados com 414 milhões de tweets publicados por 321 mil usuários. Os resultados apresentados demonstram que o uso de vida útil de tópico na tarefa de recomendação melhora a qualidade das recomendações, e o uso da versão personalizada desta informação melhorou ainda mais a qualidade destas / This document presents a study that evaluates the use of temporal information in the task of recommending tweets on Twitter. Two temporal aspects have been analysed: the lifespan of information topic and its personalized version for each user. The application of such temporal aspects has been evaluated using three recommendation systems implemented in this work. We also evaluated two topic models considered to describe tweets: a bag of words model and a model of latent topics extracted using LDA (Latent Dirichlet Allocation). Furthermore, we evaluated the use of SVM (Support Vector Machines) to estimate the user profile, comparing this approach with a simpler one. The experiments have been executed using a dataset with 414 millions of tweets published by 321 thousands of users. The results show that the use of topic lifespan information increases the quality of recommendation, and the personalized version of this information increases the quality even more

Submitted by Renata Lopes (renatasil82@gmail.com) on 2015-12-07T12:51:30Z No. of bitstreams: 1 gustavohenriquedarochareis.pdf: 1407647 bytes, checksum: b94399b5bafb2e6cb8e48443f285a7c4 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2015-12-09T10:33:11Z (GMT) No. of bitstreams: 1 gustavohenriquedarochareis.pdf: 1407647 bytes, checksum: b94399b5bafb2e6cb8e48443f285a7c4 (MD5) / Made available in DSpace on 2015-12-09T10:33:11Z (GMT). No. of bitstreams: 1 gustavohenriquedarochareis.pdf: 1407647 bytes, checksum: b94399b5bafb2e6cb8e48443f285a7c4 (MD5) Previous issue date: 2015-08-06 / O uso de dispositivos m oveis v^em aumentando signi cativamente nos ultimos anos. Outra tend^encia e a consolida c~ao no uso Tecnologias Digitais da Informa c~ao e Comunica c~ao para ns educacionais. Estes cen arios juntos possibilitam novas formas de comunica c~ao entre professores e alunos, como, por exemplo, a recomenda c~ao de conte udos educacionais e colabora c~ao utilizando dispositivos m oveis. Este trabalho mostra uma arquitetura, chamada CALearning, que re une as principais caracter sticas para um sistema de aprendizado m ovel, como promover a colabora c~ao (recomenda c~ao e avalia c~ao de conte udos) entre os alunos e fazer a recomenda c~ao de conte udos de acordo com o estilo de aprendizagem e prefer^encias do usu ario. A arquitetura tamb em faz uso de informa c~oes de contexto para recomendar conte udos adaptados de acordo com as caracter sticas de acesso a Internet (taxa de transmiss~ao) e deslocamento do aluno durante sua intera c~ao com o aplicativo. Como prova de conceito foi desenvolvido tr^es sistemas chamados CALearningDroid, CALearningWeb e CALearningWS, baseado na arquitetura proposta / The use of mobile devices have increased signi cantly in recent years. Another trend is the consolidation in using Digital Technologies of Information and Communication for educational purposes. These scenarios together enable new forms of communication between teachers and students, for example, the recommendation of educational content and collaboration using mobile devices. This work shows an architecture called CALearning, which brings together the main features for a mobile learning system as promote collaboration (recommendation and content evaluation) among students and to do the content recommendation according to the learning style and user's preferences. The architecture also does use of context information to recommend content tailored according to the Internet access features (transmission rate) and displacement of the learner during their interaction with the application. As proof of concept, was developed three systems called CALearningDroid, CALearningWeb and CALearningWS, based in the proposed architecture.

Internet Protocol TV (IPTV) delivers television content to users over IP-based network. Different from the traditional TV services, IPTV platforms provide users with large amount of multimedia contents with interactive and personalized services, including the targeted advertisement, on-demand content, personal video recorder, and so on. IPTV is promising since it allows to satisfy users experience and presents advanced entertainment services. On the other hand, the Next Generation Network (NGN) approach in allowing services convergence (through for instance coupling IPTV with the IP Multimedia Subsystem (IMS) architecture or NGN Non-IMS architecture) enhances users' experience and allows for more services personalization. Although the rapid advancement in interactive TV technology (including IPTV and NGN technologies), services personalization is still in its infancy, lacking the real distinguish of each user in a unique manner, the consideration of the context of the user (who is this user, what is his preferences, his regional area, location, ..) and his environment (characteristics of the users' devices 'screen types, size, supported resolution, '' and networks available network types to be used by the user, available bandwidth, ..') as well as the context of the service itself (content type and description, available format 'HD/SD', available language, ..) in order to provide the adequate personalized content for each user. This advanced IPTV services allows services providers to promote new services and open new business opportunities and allows network operators to make better utilization of network resources through adapting the delivered content according to the available bandwidth and to better meet the QoE (Quality of Experience) of clients. This thesis focuses on enhanced personalization for IPTV services following a user-centric context-aware approach through providing solutions for: i) Users' identification during IPTV service access through a unique and fine-grained manner (different from the identification of the subscription which is the usual current case) based on employing a personal identifier for each user which is a part of the user context information. ii) Context-Aware IPTV service through proposing a context-aware system on top of the IPTV architecture for gathering in a dynamic and real-time manner the different context information related to the user, devices, network and service. The context information is gathered throughout the whole IPTV delivery chain considering the user domain, network provider domain, and service/content provider domain. The proposed context-aware system allows monitoring user's environment (devices and networks status), interpreting user's requirements and making the user's interaction with the TV system dynamic and transparent. iii) Personalized recommendation and selection of IPTV content based on the different context information gathered and the personalization decision taken by the context-aware system (different from the current recommendation approach mainly based on matching content to users' preferences) which in turn highly improves the users' Quality of Experience (QoE) and enriching the offers of IPTV services

Le monde de la TV est en cours de transformation de la télévision analogique à la télévision numérique, qui est capable de diffuser du contenu de haute qualité, offrir aux consommateurs davantage de choix, et rendre l'expérience de visualisation plus interactive. IPTV (Internet Protocol TV) présente une révolution dans la télévision numérique dans lequel les services de télévision numérique sont fournis aux utilisateurs en utilisant le protocole Internet (IP) au dessus d’une connexion haut débit. Les progrès de la technologie IPTV permettra donc un nouveau modèle de fourniture de services. Les fonctions offertes aux utilisateurs leur permettent de plus en plus d’autonomie et de plus en plus de choix. Il en est notamment ainsi de services de type ‘nTS’ (pour ‘network Time Shifting’ en anglais) qui permettent à un utilisateur de visionner un programme de télévision en décalage par rapport à sa programmation de diffusion, ou encore des services de type ‘nPVR’ (pour ‘network Personal Video Recorder’ en anglais) qui permettent d’enregistrer au niveau du réseau un contenu numérique pour un utilisateur. D'autre part, l'architecture IMS proposée dans NGN fournit une architecture commune pour les services IPTV. Malgré les progrès rapides de la technologie de télévision interactive (comprenant notamment les technologies IPTV et NGN), la personnalisation de services IPTV en est encore à ses débuts. De nos jours, la personnalisation des services IPTV se limite principalement à la recommandation de contenus et à la publicité ciblée. Ces services ne sont donc pas complètement centrés sur l’utilisateur, alors que choisir manuellement les canaux de diffusion et les publicités désirées peut représenter une gêne pour l’utilisateur. L’adaptation des contenus numériques en fonction de la capacité des réseaux et des dispositifs utilisés n’est pas encore prise en compte dans les implémentations actuelles. Avec le développement des technologies numériques, les utilisateurs sont amenés à regarder la télévision non seulement sur des postes de télévision, mais également sur des smart phones, des tablettes digitales, ou encore des PCs. En conséquence, personnaliser les contenus IPTV en fonction de l’appareil utilisé pour regarder la télévision, en fonction des capacités du réseau et du contexte de l’utilisateur représente un défi important. Cette thèse présente des solutions visant à améliorer la personnalisation de services IPTV à partir de trois aspects: 1) Nouvelle identification et authentification pour services IPTV. 2) Nouvelle architecture IPTV intégrée et comportant un système de sensibilité au contexte pour le service de personnalisation. 3) Nouveau service de recommandation de contenu en fonction des préférences de l’utilisateur et aussi des informations contextes / Internet Protocol TV (IPTV) delivers television content to users over IP-based network. Different from the traditional TV services, IPTV platforms provide users with large amount of multimedia contents with interactive and personalized services, including the targeted advertisement, on-demand content, personal video recorder, and so on. IPTV is promising since it allows to satisfy users experience and presents advanced entertainment services. On the other hand, the Next Generation Network (NGN) approach in allowing services convergence (through for instance coupling IPTV with the IP Multimedia Subsystem (IMS) architecture or NGN Non-IMS architecture) enhances users’ experience and allows for more services personalization. Although the rapid advancement in interactive TV technology (including IPTV and NGN technologies), services personalization is still in its infancy, lacking the real distinguish of each user in a unique manner, the consideration of the context of the user (who is this user, what is his preferences, his regional area, location, ..) and his environment (characteristics of the users’ devices ‘screen types, size, supported resolution, ‘‘ and networks available network types to be used by the user, available bandwidth, ..’) as well as the context of the service itself (content type and description, available format ‘HD/SD’, available language, ..) in order to provide the adequate personalized content for each user. This advanced IPTV services allows services providers to promote new services and open new business opportunities and allows network operators to make better utilization of network resources through adapting the delivered content according to the available bandwidth and to better meet the QoE (Quality of Experience) of clients. This thesis focuses on enhanced personalization for IPTV services following a user-centric context-aware approach through providing solutions for: i) Users’ identification during IPTV service access through a unique and fine-grained manner (different from the identification of the subscription which is the usual current case) based on employing a personal identifier for each user which is a part of the user context information. ii) Context-Aware IPTV service through proposing a context-aware system on top of the IPTV architecture for gathering in a dynamic and real-time manner the different context information related to the user, devices, network and service. The context information is gathered throughout the whole IPTV delivery chain considering the user domain, network provider domain, and service/content provider domain. The proposed context-aware system allows monitoring user’s environment (devices and networks status), interpreting user’s requirements and making the user’s interaction with the TV system dynamic and transparent. iii) Personalized recommendation and selection of IPTV content based on the different context information gathered and the personalization decision taken by the context-aware system (different from the current recommendation approach mainly based on matching content to users’ preferences) which in turn highly improves the users’ Quality of Experience (QoE) and enriching the offers of IPTV services

La liste des domaines touchés par l’apprentissage machine s’allonge rapidement. Au fur et à mesure que la quantité de données disponibles augmente, le développement d’algorithmes d’apprentissage de plus en plus puissants est crucial. Ce mémoire est constitué de trois parties: d’abord un survol des concepts de bases de l’apprentissage automatique et les détails nécessaires pour l’entraînement de réseaux de neurones, modèles qui se livrent bien à des architectures profondes. Ensuite, le premier article présente une application de l’apprentissage machine aux jeux vidéos, puis une méthode de mesure performance pour ceux-ci en tant que politique de décision. Finalement, le deuxième article présente des résultats théoriques concernant l’entraînement d’architectures profondes nonsupervisées. Les jeux vidéos sont un domaine particulièrement fertile pour l’apprentissage automatique: il estf facile d’accumuler d’importantes quantités de données, et les applications ne manquent pas. La formation d’équipes selon un critère donné est une tˆache commune pour les jeux en lignes. Le premier article compare différents algorithmes d’apprentissage à des réseaux de neurones profonds appliqués à la prédiction de la balance d’un match. Ensuite nous présentons une méthode par simulation pour évaluer les modèles ainsi obtenus utilisés dans le cadre d’une politique de décision en ligne. Dans un deuxième temps nous présentons une nouvelleméthode pour entraîner des modèles génératifs. Des résultats théoriques nous indiquent qu’il est possible d’entraîner par rétropropagation des modèles non-supervisés pouvant générer des échantillons qui suivent la distribution des données. Ceci est un résultat pertinent dans le cadre de la récente littérature scientifique investiguant les propriétés des autoencodeurs comme modèles génératifs. Ces résultats sont supportés avec des expériences qualitatives préliminaires ainsi que quelques résultats quantitatifs. / The list of areas affected by machine learning is growing rapidly. As the amount of available training data increases, the development of more powerful learning algorithms is crucial. This thesis consists of three parts: first an overview of the basic concepts of machine learning and the details necessary for training neural networks, models that lend themselves well to deep architectures. The second part presents an application of machine learning to online video games, and a performance measurement method when using these models as decision policies. Finally, the third section presents theoretical results for unsupervised training of deep architectures. Video games are a particularly fertile area for machine learning: it is easy to accumulate large amounts of data, and many tasks are possible. Assembling teams of equal skill is a common machine learning application for online games. The first paper compares different learning algorithms against deep neural networks applied to the prediction of match balance in online games. We then present a simulation based method to evaluate the resulting models used as decision policies for online matchmaking. Following this we present a new method to train generative models. Theoretical results indicate that it is possible to train by backpropagation unsupervised models that can generate samples following the data’s true distribution. This is a relevant result in the context of the recent literature investigating the properties of autoencoders as generative models. These results are supported with preliminary quantitative results and some qualitative experiments.

No description available.