Papyres : un système de gestion et de recommandation d’articles de recherche

Les étudiants gradués et les professeurs (les chercheurs, en général), accèdent, passent en revue et utilisent régulièrement un grand nombre d’articles, cependant aucun des outils et solutions existants ne fournit la vaste gamme de fonctionnalités exigées pour gérer correctement ces ressources. En effet, les systèmes de gestion de bibliographie gèrent les références et les citations, mais ne parviennent pas à aider les chercheurs à manipuler et à localiser des ressources. D'autre part, les systèmes de recommandation d’articles de recherche et les moteurs de recherche spécialisés aident les chercheurs à localiser de nouvelles ressources, mais là encore échouent dans l’aide à les gérer. Finalement, les systèmes de gestion de contenu d'entreprise offrent les fonctionnalités de gestion de documents et des connaissances, mais ne sont pas conçus pour les articles de recherche. Dans ce mémoire, nous présentons une nouvelle classe de systèmes de gestion : système de gestion et de recommandation d’articles de recherche. Papyres (Naak, Hage, & Aïmeur, 2008, 2009) est un prototype qui l’illustre. Il combine des fonctionnalités de bibliographie avec des techniques de recommandation d’articles et des outils de gestion de contenu, afin de fournir un ensemble de fonctionnalités pour localiser les articles de recherche, manipuler et maintenir les bibliographies. De plus, il permet de gérer et partager les connaissances relatives à la littérature. La technique de recommandation utilisée dans Papyres est originale. Sa particularité réside dans l'aspect multicritère introduit dans le processus de filtrage collaboratif, permettant ainsi aux chercheurs d'indiquer leur intérêt pour des parties spécifiques des articles. De plus, nous proposons de tester et de comparer plusieurs approches afin de déterminer le voisinage dans le processus de Filtrage Collaboratif Multicritère, de telle sorte à accroître la précision de la recommandation. Enfin, nous ferons un rapport global sur la mise en œuvre et la validation de Papyres.

Combining collaborative and content-based filtering to recommend research papers

The number of research papers available today is growing at a staggering rate, generating a huge amount of information that people cannot keep up with. According to a tendency indicated by the United States’ National Science Foundation, more than 10 million new papers will be published in the next 20 years. Because most of these papers will be available on the Web, this research focus on exploring issues on recommending research papers to users, in order to directly lead users to papers of their interest. Recommender systems are used to recommend items to users among a huge stream of available items, according to users’ interests. This research focuses on the two most prevalent techniques to date, namely Content-Based Filtering and Collaborative Filtering. The first explores the text of the paper itself, recommending items similar in content to the ones the user has rated in the past. The second explores the citation web existing among papers. As these two techniques have complementary advantages, we explored hybrid approaches to recommending research papers. We created standalone and hybrid versions of algorithms and evaluated them through both offline experiments on a database of 102,295 papers, and an online experiment with 110 users. Our results show that the two techniques can be successfully combined to recommend papers. The coverage is also increased at the level of 100% in the hybrid algorithms. In addition, we found that different algorithms are more suitable for recommending different kinds of papers. Finally, we verified that users’ research experience influences the way users perceive recommendations. In parallel, we found that there are no significant differences in recommending papers for users from different countries. However, our results showed that users’ interacting with a research paper Recommender Systems are much happier when the interface is presented in the user’s native language, regardless the language that the papers are written. Therefore, an interface should be tailored to the user’s mother language.

Collaborative recommender agents based on case-based reasoning and trust

La comunitat científica que treballa en Intel·ligència Artificial (IA) ha dut a terme una gran quantitat de treball en com la IA pot ajudar a les persones a trobar el que volen dins d'Internet. La idea dels sistemes recomanadors ha estat extensament acceptada pels usuaris. La tasca principal d'un sistema recomanador és localitzar ítems, fonts d'informació i persones relacionades amb els interessos i preferències d'una persona o d'un grup de persones. Això comporta la construcció de models d'usuari i l'habilitat d'anticipar i predir les preferències de l'usuari. Aquesta tesi està focalitzada en l'estudi de tècniques d'IA que millorin el rendiment dels sistemes recomanadors. Inicialment, s'ha dut a terme un anàlisis detallat de l'actual estat de l'art en aquest camp. Aquest treball ha estat organitzat en forma de taxonomia on els sistemes recomanadors existents a Internet es classifiquen en 8 dimensions generals. Aquesta taxonomia ens aporta una base de coneixement indispensable pel disseny de la nostra proposta. El raonament basat en casos (CBR) és un paradigma per aprendre i raonar a partir de la experiència adequat per sistemes recomanadors degut als seus fonaments en el raonament humà. Aquesta tesi planteja una nova proposta de CBR aplicat al camp de la recomanació i un mecanisme d'oblit per perfils basats en casos que controla la rellevància i edat de les experiències passades. Els resultats experimentals demostren que aquesta proposta adapta millor els perfils als usuaris i soluciona el problema de la utilitat que pateixen el sistemes basats en CBR. Els sistemes recomanadors milloren espectacularment la qualitat dels resultats quan informació sobre els altres usuaris és utilitzada quan es recomana a un usuari concret. Aquesta tesi proposa l'agentificació dels sistemes recomanadors per tal de treure profit de propietats interessants dels agents com ara la proactivitat, la encapsulació o l'habilitat social. La col·laboració entre agents es realitza a partir del mètode de filtratge basat en la opinió i del mètode col·laboratiu de filtratge a partir de confiança. Els dos mètodes es basen en un model social de confiança que fa que els agents siguin menys vulnerables als altres quan col·laboren. Els resultats experimentals demostren que els agents recomanadors col·laboratius proposats milloren el rendiment del sistema mentre que preserven la privacitat de les dades personals de l'usuari. Finalment, aquesta tesi també proposa un procediment per avaluar sistemes recomanadors que permet la discussió científica dels resultats. Aquesta proposta simula el comportament dels usuaris al llarg del temps basat en perfils d'usuari reals. Esperem que aquesta metodologia d'avaluació contribueixi al progrés d'aquesta àrea de recerca.

Collaborative personal profiling for web service ranking and recommendation

Web service is one of the most fundamental technologies in implementing service oriented architecture (SOA) based applications. One essential challenge related to web service is to find suitable candidates with regard to web service consumer’s requests, which is normally called web service discovery. During a web service discovery protocol, it is expected that the consumer will find it hard to distinguish which ones are more suitable in the retrieval set, thereby making selection of web services a critical task. In this paper, inspired by the idea that the service composition pattern is significant hint for service selection, a personal profiling mechanism is proposed to improve ranking and recommendation performance. Since service selection is highly dependent on the composition process, personal knowledge is accumulated from previous service composition process and shared via collaborative filtering where a set of users with similar interest will be firstly identified. Afterwards a web service re-ranking mechanism is employed for personalised recommendation. Experimental studies are conduced and analysed to demonstrate the promising potential of this research.

Hierarchical graph maps for visualization of collaborative recommender systems

In this paper we provide a method that allows the visualization of similarity relationships present between items of collaborative filtering recommender systems, as well as the relative importance of each of these. The objective is to offer visual representations of the recommender system?s set of items and of their relationships; these graphs show us where the most representative information can be found and which items are rated in a more similar way by the recommender system?s community of users. The visual representations achieved take the shape of phylogenetic trees, displaying the numerical similarity and the reliability between each pair of items considered to be similar. As a case study we provide the results obtained using the public database Movielens 1M, which contains 3900 movies.

Recommendation & mobile systems - a state of the art for tourism

Recommendation systems have been growing in number over the last fifteen years. To evolve and adapt to the demands of the actual society, many paradigms emerged giving birth to even more paradigms and hybrid approaches. These approaches contain strengths and weaknesses that need to be evaluated according to the knowledge area in which the system is going to be implemented. Mobile devices have also been under an incredible growth rate in every business area, and there are already lots of mobile based systems to assist tourists. This explosive growth gave birth to different mobile applications, each having their own advantages and disadvantages. Since recommendation and mobile systems might as well be integrated, this work intends to present the current state of the art in tourism mobile and recommendation systems, as well as to state their advantages and disadvantages.

Sistema de recomendação e personalização de televisão

Com a expansão da Televisão Digital e a convergência entre os meios de difusão convencionais e a televisão sobre IP, o número de canais disponíveis tem aumentado de forma gradual colocando o espectador numa situação de difícil escolha quanto ao programa a visionar. Sobrecarregados com uma grande quantidade de programas e informação associada, muitos espectadores desistem sistematicamente de ver um programa e tendem a efectuar zapping entre diversos canais ou a assistir sempre aos mesmos programas ou canais. Diante deste problema de sobrecarga de informação, os sistemas de recomendação apresentam-se como uma solução. Nesta tese pretende estudar-se algumas das soluções existentes dos sistemas de recomendação de televisão e desenvolver uma aplicação que permita a recomendação de um conjunto de programas que representem potencial interesse ao espectador. São abordados os principais conceitos da área dos algoritmos de recomendação e apresentados alguns dos sistemas de recomendação de programas de televisão desenvolvidos até à data. Para realizar as recomendações foram desenvolvidos dois algoritmos baseados respectivamente em técnicas de filtragem colaborativa e de filtragem de conteúdo. Estes algoritmos permitem através do cálculo da similaridade entre itens ou utilizadores realizar a predição da classificação que um utilizador atribuiria a um determinado item (programa de televisão, filme, etc.). Desta forma é possível avaliar o nível de potencial interesse que o utilizador terá em relação ao respectivo item. Os conjuntos de dados que descrevem as características dos programas (título, género, actores, etc.) são armazenados de acordo com a norma TV-Anytime. Esta norma de descrição de conteúdo multimédia apresenta a vantagem de ser especificamente vocacionada para conteúdo audiovisual e está disponível livremente. O conjunto de recomendações obtidas é apresentado ao utilizador através da interacção com uma aplicação Web que permite a integração de todos os componentes do sistema. Para validação do trabalho foi considerado um dataset de teste designado de htrec2011-movielens-2k e cujo conteúdo corresponde a um conjunto de filmes classificados por diversos utilizadores num ambiente real. Este conjunto de filmes possui, para além da classificações atribuídas pelos utilizadores, um conjunto de dados que descrevem o género, directores, realizadores e país de origem. Para validação final do trabalho foram realizados diversos testes dos quais o mais relevante correspondeu à avaliação da distância entre predições e valores reais e cujo objectivo é classificar a capacidade dos algoritmos desenvolvidos preverem com precisão as classificações que os utilizadores atribuiriam aos itens analisados.

TV Recommendation and Personalization Systems: Integrating Broadcast and Video On-demand Services

he expansion of Digital Television and the convergence between conventional broadcasting and television over IP contributed to the gradual increase of the number of available channels and on demand video content. Moreover, the dissemination of the use of mobile devices like laptops, smartphones and tablets on everyday activities resulted in a shift of the traditional television viewing paradigm from the couch to everywhere, anytime from any device. Although this new scenario enables a great improvement in viewing experiences, it also brings new challenges given the overload of information that the viewer faces. Recommendation systems stand out as a possible solution to help a watcher on the selection of the content that best fits his/her preferences. This paper describes a web based system that helps the user navigating on broadcasted and online television content by implementing recommendations based on collaborative and content based filtering. The algorithms developed estimate the similarity between items and users and predict the rating that a user would assign to a particular item (television program, movie, etc.). To enable interoperability between different systems, programs characteristics (title, genre, actors, etc.) are stored according to the TV-Anytime standard. The set of recommendations produced are presented through a Web Application that allows the user to interact with the system based on the obtained recommendations.

Recommending media content based on machine learning methods

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Improving Searching and Browsing Capabilities of Learning Object Repositories

Learning object repositories are a basic piece of virtual learning environments used for content management. Nevertheless, learning objects have special characteristics that make traditional solutions for content management ine ective. In particular, browsing and searching for learning objects cannot be based on the typical authoritative meta-data used for describing content, such as author, title or publicationdate, among others. We propose to build a social layer on top of a learning object repository, providing nal users with additional services fordescribing, rating and curating learning objects from a teaching perspective. All these interactions among users, services and resources can be captured and further analyzed, so both browsing and searching can be personalized according to user pro le and the educational context, helping users to nd the most valuable resources for their learning process. In this paper we propose to use reputation schemes and collaborative filtering techniques for improving the user interface of a DSpace based learning object repository.

A RecommendationSystem for a Web Shop

Tämän työn tarkoituksena on käytännöllisen suositusjärjestelmäratkaisun kehittäminen verkkokauppaympäristöön olemassaolevaa teoriatietoa käyttäen. Työn ensimmäisessä osiossa tarkastellaan ensin tapoja lähdetiedon keräämiseksi järjestelmää varten. Tämän jälkeen käydään läpi eri menetelmiä suosituksen toteuttamiseksi. Lisäksi tutustutaan yleisiin ongelmiin eri menetelmien kanssa. Seuraavaksi tutkitaan miten järjestelmän käyttämään suositustietoa voidaan ryhmitellä. Tämänjälkeen arvioidaan esitettyjä menetelmiä yleisesti tunnettujen kriteerien perusteella. Suositusjärjestelmän toteutustyö on kuvattuna työn toisessa osiossa. Toteutettu ohjelmisto on asennettu kahteen erilliseen toimintaympäristöön.

Learning Preferences with Kernel-Based Methods

Learning of preference relations has recently received significant attention in machine learning community. It is closely related to the classification and regression analysis and can be reduced to these tasks. However, preference learning involves prediction of ordering of the data points rather than prediction of a single numerical value as in case of regression or a class label as in case of classification. Therefore, studying preference relations within a separate framework facilitates not only better theoretical understanding of the problem, but also motivates development of the efficient algorithms for the task. Preference learning has many applications in domains such as information retrieval, bioinformatics, natural language processing, etc. For example, algorithms that learn to rank are frequently used in search engines for ordering documents retrieved by the query. Preference learning methods have been also applied to collaborative filtering problems for predicting individual customer choices from the vast amount of user generated feedback. In this thesis we propose several algorithms for learning preference relations. These algorithms stem from well founded and robust class of regularized least-squares methods and have many attractive computational properties. In order to improve the performance of our methods, we introduce several non-linear kernel functions. Thus, contribution of this thesis is twofold: kernel functions for structured data that are used to take advantage of various non-vectorial data representations and the preference learning algorithms that are suitable for different tasks, namely efficient learning of preference relations, learning with large amount of training data, and semi-supervised preference learning. Proposed kernel-based algorithms and kernels are applied to the parse ranking task in natural language processing, document ranking in information retrieval, and remote homology detection in bioinformatics domain. Training of kernel-based ranking algorithms can be infeasible when the size of the training set is large. This problem is addressed by proposing a preference learning algorithm whose computation complexity scales linearly with the number of training data points. We also introduce sparse approximation of the algorithm that can be efficiently trained with large amount of data. For situations when small amount of labeled data but a large amount of unlabeled data is available, we propose a co-regularized preference learning algorithm. To conclude, the methods presented in this thesis address not only the problem of the efficient training of the algorithms but also fast regularization parameter selection, multiple output prediction, and cross-validation. Furthermore, proposed algorithms lead to notably better performance in many preference learning tasks considered.

Informed recommender: basing recommendations on consumer product reviews

Recommender systems attempt to predict items in which a user might be interested, given some information about the user's and items' profiles. Most existing recommender systems use content-based or collaborative filtering methods or hybrid methods that combine both techniques (see the sidebar for more details). We created Informed Recommender to address the problem of using consumer opinion about products, expressed online in free-form text, to generate product recommendations. Informed recommender uses prioritized consumer product reviews to make recommendations. Using text-mining techniques, it maps each piece of each review comment automatically into an ontology

Algorithmes de recommandation musicale

Ce mémoire est composé de trois articles qui s’unissent sous le thème de la recommandation musicale à grande échelle. Nous présentons d’abord une méthode pour effectuer des recommandations musicales en récoltant des étiquettes (tags) décrivant les items et en utilisant cette aura textuelle pour déterminer leur similarité. En plus d’effectuer des recommandations qui sont transparentes et personnalisables, notre méthode, basée sur le contenu, n’est pas victime des problèmes dont souffrent les systèmes de filtrage collaboratif, comme le problème du démarrage à froid (cold start problem). Nous présentons ensuite un algorithme d’apprentissage automatique qui applique des étiquettes à des chansons à partir d’attributs extraits de leur fichier audio. L’ensemble de données que nous utilisons est construit à partir d’une très grande quantité de données sociales provenant du site Last.fm. Nous présentons finalement un algorithme de génération automatique de liste d’écoute personnalisable qui apprend un espace de similarité musical à partir d’attributs audio extraits de chansons jouées dans des listes d’écoute de stations de radio commerciale. En plus d’utiliser cet espace de similarité, notre système prend aussi en compte un nuage d’étiquettes que l’utilisateur est en mesure de manipuler, ce qui lui permet de décrire de manière abstraite la sorte de musique qu’il désire écouter.

Mures : Un système de recommandation de musique

Pendant la dernière décennie nous avons vu une transformation incroyable du monde de la musique qui est passé des cassettes et disques compacts à la musique numérique en ligne. Avec l'explosion de la musique numérique, nous avons besoin de systèmes de recommandation de musique pour choisir les chansons susceptibles d’être appréciés à partir de ces énormes bases de données en ligne ou personnelles. Actuellement, la plupart des systèmes de recommandation de musique utilisent l’algorithme de filtrage collaboratif ou celui du filtrage à base de contenu. Dans ce mémoire, nous proposons un algorithme hybride et original qui combine le filtrage collaboratif avec le filtrage basé sur étiquetage, amélioré par la technique de filtrage basée sur le contexte d’utilisation afin de produire de meilleures recommandations. Notre approche suppose que les préférences de l'utilisateur changent selon le contexte d'utilisation. Par exemple, un utilisateur écoute un genre de musique en conduisant vers son travail, un autre type en voyageant avec la famille en vacances, un autre pendant une soirée romantique ou aux fêtes. De plus, si la sélection a été générée pour plus d'un utilisateur (voyage en famille, fête) le système proposera des chansons en fonction des préférences de tous ces utilisateurs. L'objectif principal de notre système est de recommander à l'utilisateur de la musique à partir de sa collection personnelle ou à partir de la collection du système, les nouveautés et les prochains concerts. Un autre objectif de notre système sera de collecter des données provenant de sources extérieures, en s'appuyant sur des techniques de crawling et sur les flux RSS pour offrir des informations reliées à la musique tels que: les nouveautés, les prochains concerts, les paroles et les artistes similaires. Nous essayerons d’unifier des ensembles de données disponibles gratuitement sur le Web tels que les habitudes d’écoute de Last.fm, la base de données de la musique de MusicBrainz et les étiquettes des MusicStrands afin d'obtenir des identificateurs uniques pour les chansons, les albums et les artistes.