Submitted by Erika Demachki (erikademachki@gmail.com) on 2015-03-24T21:17:09Z No. of bitstreams: 3 Dissertação - Luiz Mario Lustosa Pascoal - 2014.pdf: 7280181 bytes, checksum: 68a6ac0602e3e51f6e6952bbd6916150 (MD5) FunctionApproximator.zip: 2288624 bytes, checksum: 178c2e6a0b080b3d0548836974016236 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Erika Demachki (erikademachki@gmail.com) on 2015-03-24T21:19:16Z (GMT) No. of bitstreams: 3 Dissertação - Luiz Mario Lustosa Pascoal - 2014.pdf: 7280181 bytes, checksum: 68a6ac0602e3e51f6e6952bbd6916150 (MD5) FunctionApproximator.zip: 2288624 bytes, checksum: 178c2e6a0b080b3d0548836974016236 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-03-24T21:19:16Z (GMT). No. of bitstreams: 3 Dissertação - Luiz Mario Lustosa Pascoal - 2014.pdf: 7280181 bytes, checksum: 68a6ac0602e3e51f6e6952bbd6916150 (MD5) FunctionApproximator.zip: 2288624 bytes, checksum: 178c2e6a0b080b3d0548836974016236 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2014-08-22 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / With the development of web 2.0, social networks have achieved great space on the internet, with that many users provide information and interests about themselves. There are expert systems that make use of the user’s interests to recommend different products, these systems are known as Recommender Systems. One of the main techniques of a Recommender Systems is the Collaborative Filtering (User-based) which recommends products to users based on what other similar people liked in the past. Therefore, this work presents model approximation of functions that generates rankings, that through a Genetic Algorithm, is able to learn an approximation function composed by different social variables, customized for each Facebook user. The learned function must be able to reproduce a ranking of people (friends) originally created with user’s information, that apply some influence in the user’s decision. As a case study, this work discusses the context of events through information regarding the frequency of participation of some users at several distinct events. Two different approaches on learning and applying the approximation function have been developed. The first approach provides a general model that learns a function in advance and then applies it in a set of test data and the second approach presents an specialist model that learns a specific function for each test scenario. Two proposals for evaluating the ordering created by the learned function, called objective functions A and B, where the results for both objective functions show that it is possible to obtain good solutions with the generalist and the specialist approaches of the proposed method. / Com o desenvolvimento da Web 2.0, as redes sociais têm conquistado grande espaço na internet, com isso muitos usuários acabam fornecendo diversas informações e interesses sobre si mesmos. Existem sistemas especialistas que fazem uso dos interesses do usuário para recomendar diferentes produtos, esses sistemas são conhecidos como Sistemas de Recomendação. Uma das principais técnicas de um Sistema de Recomendação é a Filtragem Colaborativa (User-based) que recomenda produtos para seus usuários baseados no que outras pessoas similares à ele tenham gostado no passado. Portanto, este trabalho apresenta um modelo de aproximação de funções geradora de rankings que, através de um Algoritmo Genético, é capaz de aprender uma função de aproximação composta por diferentes atributos sociais, personalizada para cada usuário do Facebook. A função aprendida deve ser capaz de reproduzir um ranking de pessoas (amigos) criado originalmente com informações do usuário, que exercem certa influência na decisão do usuário. Como estudo de caso, esse trabalho aborda o contexto de eventos através de informações com relação a frequência de participação de alguns usuários em vários eventos distintos. Foram desenvolvidas duas abordagens distintas para aprendizagem e aplicação da função de aproximação. A primeira abordagem apresenta um modelo generalista, que previamente aprende uma função e em seguida a aplica em um conjunto de dados de testes e a segunda abordagem apresenta um modelo especialista, que aprende uma função específica para cada cenário de teste. Também foram apresentadas duas propostas para avaliação da ordenação criada pela função aprendida, denominadas funções objetivo A e B, onde os resultados para ambas as funções objetivo A e B mostram que é possível obter boas soluções com as abordagens generalista e especialista do método proposto.

The goal of this project was to create a web-based, crowd-sourced, correlational database, that easily allowed users to submit objects and receive correlated objects as results. The webservice was created in the web development languages of HTML, CSS, PHP and Javscript, with MySQL to handle the database. Simultaneous development was kept in check with the aid of the source code management system GIT. Upon completion, the service contained several HTML-views, the ability to add and rate objects, a per-object dedicated page with information parsed from Wikipedia.org, and a view with objects ranked in accordance to the preferences specific to the current user. Roughly a month after the beginning of development, the website was publicly launched and promoted in order to collect data, and improvements were added to the website as needed. Two weeks after the public launch, the collected data was measured and analyzed. The algorithm proved effective and scalable, especially with the introduction of tags and simultaneous computation of object features.

A crescente oferta de programas de código aberto na rede mundial de computadores expõe potenciais usuários a muitas possibilidades de escolha. Em face da pluralidade de interesses desses indivíduos, mecanismos eficientes que os aproximem daquilo que buscam trazem benefícios para eles próprios, assim como para os desenvolvedores dos programas. Este trabalho apresenta o AppRecommender, um recomendador de aplicativos GNU/Linux que realiza uma filtragem no conjunto de programas disponíveis e oferece sugestões individualizadas para os usuários. Tal feito é alcançado por meio da análise de perfis e descoberta de padrões de comportamento na população estudada, de sorte que apenas os aplicativos considerados mais suscetíveis a aceitação sejam oferecidos aos usuários. / The increasing availability of open source software on the World Wide Web exposes potential users to a wide range of choices. Given the individuals plurality of interests, mechanisms that get them close to what they are looking for would benefit users and software developers. This work presents AppRecommender, a recommender system for GNU/Linux applications which performs a filtering on the set of available software and individually offers suggestions to users. This is achieved by analyzing profiles and discovering patterns of behavior of the studied population, in a way that only those applications considered most prone to acceptance are presented to users.

The goal of this project is to explore the problem of product recommendations in the area of e-commerce and to evaluate known techniques, design product recommendation system for an existing e-commerce site, implement it and test it. This report introduces the problem, briefly examines current state of affairs in this area and defines requirements for a product recommendation module. The concept of data mining in general is introduced. The report proceeds to present detailed design corresponding to defined requirements and summarizes data gathered during testing phase. It concludes with evaluation and with discussion of the remaining goals for this thesis.

This master's thesis describes creation of recommender system that is used in real server cbdb.cz. A~fully operational recommender system was developed using collaborative and content-based filtering techniques. Thanks to many user feedback, we were able to evaluate their opinion. Many recommended books were tagged as desirable. This thesis is extending current functionality of cbdb.cz with recommender system. This system uses its extensive database of ratings, users and books.

This thesis deals with problems of recommender systems and their usage in web applications. There are three main data mining techniques summarized and individual approaches for recommendation. Main part of this thesis is a suggestion and an implementation of web applications for recommending dishes from restaurants. Algorithm for food recommending is designed and implemented in this paper. The algorithm deals with the problem of frequently changing items. The algorithm utilizes hybrid filtering technique which is based on content and knowledge. This filtering technique uses cosine vector similarity for computation.

Notre recherche porte sur la recommandation de nouvelles offres d'emploi venant d'être postées et n'ayant pas d'historique d'interactions (démarrage à froid). Nous adaptons les systèmes de recommandations bien connus dans le domaine du commerce électronique à cet objectif, en exploitant les traces d'usage de l'ensemble des demandeurs d'emploi sur les offres antérieures. Une des spécificités du travail présenté est d'avoir considéré des données réelles, et de s'être attaqué aux défis de l'hétérogénéité et du bruit des documents textuels. La contribution présentée intègre l'information des données collaboratives pour apprendre une nouvelle représentation des documents textes, requise pour effectuer la recommandation dite à froid d'une offre nouvelle. Cette représentation dite latente vise essentiellement à construire une bonne métrique. L'espace de recherche considéré est celui des réseaux neuronaux. Les réseaux neuronaux sont entraînés en définissant deux fonctions de perte. La première cherche à préserver la structure locale des informations collaboratives, en s'inspirant des approches de réduction de dimension non linéaires. La seconde s'inspire des réseaux siamois pour reproduire les similarités issues de la matrice collaborative. Le passage à l'échelle de l'approche et ses performances reposent sur l'échantillonnage des paires d'offres considérées comme similaires. L'intérêt de l'approche proposée est démontrée empiriquement sur les données réelles et propriétaires ainsi que sur le benchmark publique CiteULike. Enfin, l'intérêt de la démarche suivie est attesté par notre participation dans un bon rang au challenge international RecSys 2017 (15/100; un million d'utilisateurs pour un million d'offres). / Our research focuses on the recommendation of new job offers that have just been posted and have no interaction history (cold start). To this objective, we adapt well-knowns recommendations systems in the field of e-commerce by exploiting the record of use of all job seekers on previous offers. One of the specificities of the work presented is to have considered real data, and to have tackled the challenges of heterogeneity and noise of textual documents. The presented contribution integrates the information of the collaborative data to learn a new representation of text documents, which is required to make the so-called cold start recommendation of a new offer. The new representation essentially aims to build a good metric. The search space considered is that of neural networks. Neural networks are trained by defining two loss functions. The first seeks to preserve the local structure of collaborative information, drawing on non-linear dimension reduction approaches. The second is inspired by Siamese networks to reproduce the similarities from the collaborative matrix. The scaling up of the approach and its performance are based on the sampling of pairs of offers considered similar. The interest of the proposed approach is demonstrated empirically on the real and proprietary data as well as on the CiteULike public benchmark. Finally, the interest of the approach followed is attested by our participation in a good rank in the international challenge RecSys 2017 (15/100, with millions of users and millions of offers).

[pt] Esta tese propõe que a combinação de técnicas de filtragem colaborativa, em particular para recomendações item-item, com novas tecnologias de computação em nuvem, pode melhorar drasticamente a eficiência dos sistemas de recomendação, particularmente em situações em que o número de itens e usuários supera milhões de objetos. Nela apresentamos uma arquitetura de recomendação item-item em tempo real, que racionaliza o uso dos recursos computacionais através da computação sob demanda. A arquitetura proposta oferece uma solução para o cálculo de similaridade entre itens em tempo real, sem ter que recorrer à simplificação do modelo de recomendação ou o uso de amostragem de dados de entrada. Esta tese também apresenta um novo modelo de feedback implícito para vídeos de curta duração, que se adapta ao comportamento dos usuários, e descreve como essa arquitetura foi usada na implementação de um sistema de recomendação de vídeo em uso pelo maior grupo de mídia da América Latina, apresentando resultados de um estudo de caso real para mostrar que é possível reduzir drasticamente o tempo de cálculo das recomendações (e os custos financeiros globais) usando o provisionamento dinâmico de recursos na nuvem. Ela discute ainda a implementação em detalhes, em particular o projeto da arquitetura baseada em nuvem. Finalmente, ela também apresenta oportunidades de pesquisa em potencial que surgem a partir desta mudança de paradigma. / [en] This dissertation argues that the combination of collaborative filtering techniques, particularly for item-item recommendations, with emergent cloud computing technology can drastically improve algorithm efficiency, particularly in situations where the number of items and users scales up to several million objects. It introduces a real-time item-item recommendation architecture, which rationalizes the use of resources by exploring on-demand computing. The proposed architecture provides a real-time solution for computing online item similarity, without having to resort to either model simplification or the use of input data sampling. This dissertation also presents a new adaptive model for implicit user feedback for short videos, and describes how this architecture was used in a large scale implementation of a video recommendation system in use by the largest media group in Latin America, presenting results from a real life case study to show that it is possible to greatly reduce recommendation times (and overall financial costs) by using dynamic resource provisioning in the Cloud. It discusses the implementation in detail, in particular the design of cloud based features. Finally, it also presents potential research opportunities that arise from this paradigm shift.

In our daily life, time is of the essence. People do not have time to browse through hundreds of thousands of digital items every day to find the right item for them. This is where a recommendation system shines. Tigerhall is a company that distributes podcasts, ebooks and events to subscribers. They are expanding their digital content warehouse which leads to more data for the users to filter. To make it easier for users to find the right podcast or the most exciting e-book or event, a recommendation system has been implemented. A recommender system can be implemented in many different ways. There are content-based filtering methods that can be used that focus on information about the items and try to find relevant items based on that. Another alternative is to use collaboration filtering methods that use information about what the consumer has previously consumed in correlation with what other users have consumed to find relevant items. In this project, a hybrid recommender system that uses a k-nearest neighbors algorithm alongside a matrix factorization algorithm has been implemented. The k-nearest neighbors algorithm performed well despite the sparse data while the matrix factorization algorithm performs worse. The matrix factorization algorithm performed well when the user has consumed plenty of items.

Les systèmes de recommandation visent à présélectionner et présenter en premier les informations susceptibles d'intéresser les utilisateurs. Ceci a suscité l'attention du commerce électronique, où l'historique des achats des utilisateurs sont analysés pour prédire leurs intérêts futurs et pouvoir personnaliser les offres ou produits (appelés aussi items) qui leur sont proposés. Dans ce cadre, les systèmes de recommandation exploitent les préférences des utilisateurs et les caractéristiques des produits et des utilisateurs pour prédire leurs préférences pour des futurs items. Bien qu'ils aient démontré leur précision, ces systèmes font toujours face à de grands défis tant pour le monde académique que pour l'industrie : ces techniques traitent un grand volume de données qui exige une parallélisation des traitements, les données peuvent être également très hétérogènes, et les systèmes de recommandation souffrent du démarrage à froid, situation dans laquelle le système n'a pas (ou pas assez) d'informations sur (les nouveaux) utilisateurs/items pour proposer des recommandations précises. La technique de factorisation matricielle a démontré une précision dans les prédictions et une simplicité de passage à l'échelle. Cependant, cette approche a deux inconvénients : la complexité d'intégrer des données hétérogènes externes (telles que les caractéristiques des items) et le démarrage à froid pour un nouvel utilisateur. Cette thèse a pour objectif de proposer un système offrant une précision dans les recommandations, un passage à l'échelle pour traiter des données volumineuses, et permettant d'intégrer des données variées sans remettre en question l'indépendance du système par rapport au domaine d'application. De plus, le système doit faire face au démarrage à froid utilisateurs car il est important de fidéliser et satisfaire les nouveaux utilisateurs. Cette thèse présente quatre contributions au domaine des systèmes de recommandation: (1) nous proposons une implémentation d'un algorithme de recommandation de factorisation matricielle parallélisable pour assurer un meilleur passage à l'échelle, (2) nous améliorons la précision des recommandations en prenant en compte l'intérêt implicite des utilisateurs dans les attributs des items, (3) nous proposons une représentation compacte des caractéristiques des utilisateurs/items basée sur les filtres de bloom permettant de réduire la quantité de mémoire utile, (4) nous faisons face au démarrage à froid d'un nouvel utilisateur en utilisant des techniques d'apprentissage actif. La phase d'expérimentation utilise le jeu de données MovieLens et la base de données IMDb publiquement disponibles, ce qui permet d'effectuer des comparaisons avec des techniques existantes dans l'état de l'art. Ces expérimentations ont démontré la précision et l'efficacité de nos approches. / Recommender Systems aim at pre-selecting and presenting first the information in which users may be interested. This has raised the attention of the e-commerce, where the interests of users are analysed in order to predict future interests and to personalize the offers (a.k.a. items). Recommender systems exploit the current preferences of users and the features of items/users in order to predict their future preference in items.Although they demonstrate accuracy in many domains, these systems still face great challenges for both academia and industry: they require distributed techniques to deal with a huge volume of data, they aim to exploit very heterogeneous data, and they suffer from cold-start, situation in which the system has not (enough) information about (new) users/items to provide accurate recommendations. Among popular techniques, Matrix Factorization has demonstrated high accurate predictions and scalability to parallelize the analysis among multiple machines. However, it has two main drawbacks: (1) difficulty of integrating external heterogeneous data such as items' features, and (2) the cold-start issue. The objective of this thesis is to answer to many challenges in the field of recommender systems: (1) recommendation techniques deal with complex analysis and a huge volume of data; in order to alleviate the time consumption of analysis, these techniques need to parallelize the process among multiple machines, (2) collaborative filtering techniques do not naturally take into account the items' descriptions in the recommendation, although this information may help to perform more accurate recommendations, (3) users' and items' descriptions in very large dataset contexts can become large and memory-consuming; this makes data analysis more complex, and (4) the new user cold-start is particularly important to perform new users' recommendations and to assure new users fidelity. Our contributions to this area are given by four aspects: (1) we improve the distribution of a matrix factorization recommendation algorithm in order to achieve better scalability, (2) we enhance recommendations performed by matrix factorization by studying the implicit interest of the users in the attributes of the items, (3) we propose an accurate and low-space binary vector based on Bloom Filters for representing users/items through a high quantity of features in low memory-consumption, and (4) we cope with the new user cold-start in collaborative filtering by using active learning techniques. The experimentation phase uses the publicly available MovieLens dataset and IMDb database, what allows to perform fair comparisons to the state of the art. Our contributions demonstrate their performance in terms of accuracy and efficiency.