Orientadores: Ricardo da Silva Torres, Anderson de Rezende Rocha / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-25T19:48:27Z (GMT). No. of bitstreams: 1 Werneck_RafaeldeOliveira_M.pdf: 2395829 bytes, checksum: 8f190aeb6dbafb841d0c03f7d7099041 (MD5) Previous issue date: 2014 / Resumo: Devido ao grande crescimento do uso de tecnologias para a aquisição de dados, temos que lidar com grandes e complexos conjuntos de dados a fim de extrair conhecimento que possa auxiliar o processo de tomada de decisão em diversos domínios de aplicação. Uma solução típica para abordar esta questão se baseia na utilização de métodos de aprendizado de máquina, que são métodos computacionais que extraem conhecimento útil a partir de experiências para melhorar o desempenho de aplicações-alvo. Existem diversas bibliotecas e arcabouços na literatura que oferecem apoio à execução de experimentos de aprendizado de máquina, no entanto, alguns não são flexíveis o suficiente para poderem ser estendidos com novos métodos, além de não oferecerem mecanismos que permitam o reuso de soluções de sucesso concebidos em experimentos anteriores na ferramenta. Neste trabalho, propomos um arcabouço para automatizar experimentos de aprendizado de máquina, oferecendo um ambiente padronizado baseado em workflow, tornando mais fácil a tarefa de avaliar diferentes descritores de características, classificadores e abordagens de fusão em uma ampla gama de tarefas. Também propomos o uso de medidas de similaridade e métodos de learning-to-rank em um cenário de recomendação, para que usuários possam ter acesso a soluções alternativas envolvendo experimentos de aprendizado de máquina. Nós realizamos experimentos com quatro medidas de similaridade (Jaccard, Sorensen, Jaro-Winkler e baseada em TF-IDF) e um método de learning-to-rank (LRAR) na tarefa de recomendar workflows modelados como uma sequência de atividades. Os resultados dos experimentos mostram que a medida Jaro-Winkler obteve o melhor desempenho, com resultados comparáveis aos observados para o método LRAR. Em ambos os casos, as recomendações realizadas são promissoras, e podem ajudar usuários reais em diferentes tarefas de aprendizado de máquina / Abstract: Due to the large growth of the use of technologies for data acquisition, we have to handle large and complex data sets in order to extract knowledge that can support the decision-making process in several domains. A typical solution for addressing this issue relies on the use of machine learning methods, which are computational methods that extract useful knowledge from experience to improve performance of target applications. There are several libraries and frameworks in the literature that support the execution of machine learning experiments. However, some of them are not flexible enough for being extended with novel methods and they do not support reusing of successful solutions devised in previous experiments made in the framework. In this work, we propose a framework for automating machine learning experiments that provides a workflow-based standardized environment and makes it easy to evaluate different feature descriptors, classifiers, and fusion approaches in a wide range of tasks. We also propose the use of similarity measures and learning-to-rank methods in a recommendation scenario, in which users may have access to alternative machine learning experiments. We performed experiments with four similarity measures (Jaccard, Sorensen, Jaro-Winkler, and a TF-IDF-based measure) and one learning-to-rank method (LRAR) in the task of recommending workflows modeled as a sequence of activities. Experimental results show that Jaro-Winkler yields the highest effectiveness performance with comparable results to those observed for LRAR. In both cases, the recommendations performed are very promising and might help real-world users in different daily machine learning tasks / Mestrado / Ciência da Computação / Mestre em Ciência da Computação

Orientador: Stanley Robson de Medeiros Oliveira / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Agrícola / Made available in DSpace on 2018-08-23T12:56:12Z (GMT). No. of bitstreams: 1 Barros_FlavioMargaritoMartinsde_M.pdf: 2098709 bytes, checksum: 4fad46ce03410953cd3fbac10f9a43bd (MD5) Previous issue date: 2013 / Resumo: Sistemas de informação web oferecem informações em quantidade elevada, tal que a tarefa de encontrar a informação de interesse torna-se desafiadora. A Agencia de Informação Embrapa e um sistema web com o objetivo de organizar, tratar, armazenar e divulgar informações técnicas e conhecimentos gerados pela EMBRAPA (Empresa Brasileira de Pesquisa Agropecuária). O portal esta estruturado como uma arvore hierárquica, denominada Arvore de Conhecimento, a qual compreende centenas de paginas web, artigos, planilhas e materiais multimídia. Diariamente o site recebe milhares de acessos tal que os registros dessas visitas são armazenados em um banco de dados. Em domínios onde estão disponíveis informações em quantidade elevada, armazenadas em bancos de dados, as ferramentas de Mineração de Dados são promissoras, pois apresentam recursos para analise e extração de padrões de uso do site para fazer recomendações. Recomendações personalizadas de conteúdo melhoram a usabilidade de sistemas, agregam valor aos serviços, poupam tempo e fidelizam usuários. O objetivo desse trabalho foi projetar, desenvolver e implantar um sistema de recomendação web, baseado em regras de associação, que ofereça recomendações automaticamente de conteúdos da cultura da cana-de-açúcar, de acordo com o perfil da comunidade de usuários. Os dados utilizados nessa pesquisa foram extraídos de um banco de dados de acessos do projeto Agencia de Informação Embrapa. A metodologia utilizada na pesquisa compreendeu a preparação dos dados de visitas ao site para uma estrutura de "lista de acessos", onde estão registradas todas as paginas visitadas por cada usuário. A partir destas listas de acesso, regras de associação entre paginas foram geradas por meio do algoritmo Apriori. O conjunto de regras deu origem a uma base de conhecimento que foi armazenada em um banco de dados para fazer recomendações de conteúdo aos usuários. Como suporte a base de conhecimento, para cada pagina da agencia cana-de-açúcar foi criada uma lista de ate três das paginas mais visitadas. Essas paginas podem ser oferecidas caso haja ausência de recomendações. O sistema de recomendação foi avaliado com uma métrica denominada taxa de rejeição e, por meio de um questionário aplicado a um conjunto de usuários, foi avaliada a usabilidade da Agencia cana-de-açúcar, apos a implantação do sistema. A base de conhecimento, gerada na forma de regras de recomendação, também foi avaliada em relação a estrutura de links da Agencia, para verificar se a lista de recomendações trouxe conhecimentos sobre a estrutura do portal. De acordo com os resultados da pesquisa, por meio das recomendações, usuários encontram informações relevantes associadas as suas visitas, aumentam seu tempo de permanência no site e aumentam o uso e visualização dos conteúdos da Agencia de Informação Embrapa - Arvore cana-de-açúcar. Em paginas com dezenas de links, a base de conhecimento também atua como uma forma de resumo, apontando os principais links nas paginas / Abstract: Web information systems provide a great amount of information, so that the task of retrieving the information of interest becomes a challenge. Embrapa Information Agency is a web system aimed to organize, treat, store and disseminate technical information and knowledge generated by EMBRAPA (Brazilian Agricultural Research Corporation). The Agency's portal is structured as a hierarchical tree, called Knowledge Tree, which comprises hundreds of web pages, articles, spreadsheets and multimedia materials. Everyday this site receives thousands of access and the records of these visits are stored in a database. In domains where information is available in high quantity, stored in databases, Data Mining tools are promising, since they have resources for extraction and analysis of usage patterns of the site to make recommendations. Personalized recommendations of content improve the usability of systems, add value to services, save time and retain users. The aim of this work was to design, develop and deploy a web recommendation system based on association rules, which offers automatically recommendations of sugarcane contents, according to the profile of user community. The data used in this study were extracted from a database of accesses from Embrapa Information Agency. The methodology used in the research included a data preparation procedure to transform website visits into a structured access list, in which all page views by each user are stored. From these access lists, association rules between pages were generated by means of the Apriori algorithm. The set of rules has created a knowledge base that was stored in a database to make content recommendations to users. To support the knowledge base, for each page of the sugarcane Agency was created a list of up to three of the most visited pages. These pages can be offered if there are no recommendations. The recommender system was evaluated by using a metric called bounce rate. In addition, through a questionnaire applied to a set of users, the usability of the sugarcane Agency was evaluated, after the system deployment. The knowledge base generated in the form of recommendation rules was also evaluated in relation to link structure of Agency, to verify if the list of recommendations brought knowledge about the structure of the portal. According to the survey results, users find relevant information associated with their visits, increase their time spent on the site and increase the use and the interest of the contents of sugarcane Agency. In pages with dozens of links, the knowledge base also acts as a form of summarizing them, indicating the main links on the pages / Mestrado / Planejamento e Desenvolvimento Rural Sustentável / Mestre em Engenharia Agrícola

[pt] Com a quantidade de informação e sua disponibilidade facilitada pelo uso da Internet, diversas opções são oferecidas às pessoas e estas, normalmente, possuem pouca ou quase nenhuma experiência para decidir dentre as alternativas existentes. Neste âmbito, os Sistemas de Recomendação surgem para organizar e recomendar automaticamente, através de Aprendizado de Máquina, itens interessantes aos usuários. Um dos grandes desafios deste tipo de sistema é realizar o casamento correto entre o que está sendo recomendado e aqueles que estão recebendo a recomendação. Este trabalho aborda um Sistema de Recomendação baseado em Filtragem Colaborativa, técnica cuja essência está na troca de experiências entre usuários com interesses comuns. Na Filtragem Colaborativa, os usuários pontuam cada item experimentado de forma a indicar sua relevância, permitindo que outros do mesmo grupo se beneficiem destas pontuações. Nosso objetivo é utilizar um algoritmo de Boosting para otimizar a performance dos Sistemas de Recomendação. Para isto, utilizamos uma base de dados de anúncios com fins de validação e uma base de dados de filmes com fins de teste. Após adaptações nas estratégias convencionais de Boosting, alcançamos melhorias de até 3% sobre a performance do algoritmo original. / [en] With the amount of information and its easy availability on the Internet, many options are offered to the people and they, normally, have little or almost no experience to decide between the existing alternatives. In this scene, the Recommendation Systems appear to organize and recommend automatically, through Machine Learning, the interesting items. One of the great recommendation challenges is to match correctly what is being recommended and who are receiving the recommendation. This work presents a Recommendation System based on Collaborative Filtering, technique whose essence is the exchange of experiences between users with common interests. In Collaborative Filtering, users rate each experimented item indicating its relevance allowing the use of ratings by other users of the same group. Our objective is to implement a Boosting algorithm in order to optimize a Recommendation System performance. For this, we use a database of advertisements with validation purposes and a database of movies with testing purposes. After adaptations in the conventional Boosting strategies, improvements of 3% were reached over the original algorithm.

Social networking has encouraged users to find new ways to create, post, search, collaborate and share information of various forms. Unfortunately there is a lot of data in social networks that is not well-managed, which makes the experience within these networks less than optimal. Therefore people generally need more and more time as well as advanced tools that are used for seeking relevant information. A new search paradigm is emerging, where the user perspective is completely reversed: from finding to being found. The aim of present thesis research is to evaluate two approaches of identifying content of interest: user-initiated and system-initiated. The most suitable approaches will be implemented. Various recommendation systems for system-initiated content recommendations will also be investigated, and the best suited ones implemented. The analysis that was performed demonstrated that the users have used all of the implemented approaches and have provided positive and negative comments for all of them, which reinforces the belief that the methods for the implementation were selected correctly. The results of the user testing of the methods were evaluated based on the amount of time it took the users to find the desirable content and on the correspondence of the result compared to the user expectations.

Notre travail concerne les systèmes d’aide à la visite de musée et l’accès au patrimoine culturel. L’objectif est de concevoir des systèmes de recommandation, implémentés sur dispositifs mobiles, pour améliorer l’expérience du visiteur, en lui recommandant les items les plus pertinents et en l’aidant à personnaliser son parcours. Nous considérons essentiellement deux terrains d’application : la visite de musées et le tourisme. Nous proposons une approche de recommandation hybride et sensible au contexte qui utilise trois méthodes différentes : démographique, sémantique et collaborative. Chaque méthode est adaptée à une étape spécifique de la visite de musée. L’approche démographique est tout d’abord utilisée afin de résoudre le problème du démarrage à froid. L’approche sémantique est ensuite activée pour recommander à l’utilisateur des œuvres sémantiquement proches de celles qu’il a appréciées. Enfin l’approche collaborative est utilisée pour recommander à l’utilisateur des œuvres que les utilisateurs qui lui sont similaires ont aimées. La prise en compte du contexte de l’utilisateur se fait à l’aide d’un post-filtrage contextuel, qui permet la génération d’un parcours personnalisé dépendant des œuvres qui ont été recommandées et qui prend en compte des informations contextuelles de l’utilisateur à savoir : l’environnement physique, la localisation ainsi que le temps de visite. Dans le domaine du tourisme, les points d’intérêt à recommander peuvent être de différents types (monument, parc, musée, etc.). La nature hétérogène de ces points d’intérêt nous a poussé à proposer un système de recommandation composite. Chaque recommandation est une liste de points d’intérêt, organisés sous forme de packages, pouvant constituer un parcours de l’utilisateur. L’objectif est alors de recommander les Top-k packages parmi ceux qui satisfont les contraintes de l’utilisateur (temps et coût de visite par exemple). Nous définissons une fonction de score qui évalue la qualité d’un package suivant trois critères : l’appréciation estimée de l’utilisateur, la popularité des points d’intérêt ainsi que la diversité du package et nous proposons un algorithme inspiré de la recherche composite pour construire la liste des packages recommandés. L’évaluation expérimentale du système que nous avons proposé, en utilisant un data-set réel extrait de Tripadvisor démontre sa qualité et sa capacité à améliorer à la fois la précision et la diversité des recommandations. / Our work concerns systems that help users during museum visits and access to cultural heritage. Our goal is to design recommender systems, implemented in mobile devices to improve the experience of the visitor, by recommending him the most relevant items and helping him to personalize the tour he makes. We consider two mainly domains of application : museum visits and tourism. We propose a context-aware hybrid recommender system which uses three different methods : demographic, semantic and collaborative. Every method is adapted to a specific step of the museum tour. First, the demographic approach is used to solve the problem of the cold start. The semantic approach is then activated to recommend to the user artworks that are semantically related to those that the user appreciated. Finally, the collaborative approach is used to recommend to the user artworks that users with similar preferences have appreciated. We used a contextual post filtering to generate personalized museum routes depending on artworks which were recommended and contextual information of the user namely : the physical environment, the location as well as the duration of the visit. In the tourism field, the items to be recommended can be of various types (monuments, parks, museums, etc.). Because of the heterogeneous nature of these points of interest, we proposed a composite recommender system. Every recommendation is a list of points of interest that are organized in a package, where each package may constitute a tour for the user. The objective is to recommend the Top-k packages among those who satisfy the constraints of the user (time, cost, etc.). We define a scoring function which estimates the quality of a package according to three criteria : the estimated appreciation of the user, the popularity of points of interest as well as the diversity of packages. We propose an algorithm inspired by composite retrieval to build the list of recommended packages. The experimental evaluation of the system we proposed using a real world data set crawled from Tripadvisor demonstrates its quality and its ability to improve both the relevance and the diversity of recommendations.

La gestion efficace de grandes quantités d’informations est devenue un défi de plus en plus importantpour les systèmes d’information. Tous les jours, de nouvelles sources d’informations émergent surle web. Un humain peut assez facilement retrouver ce qu’il cherche, lorsqu’il s’agit d’un article,d’une vidéo, d’un artiste précis. En revanche, il devient assez difficile, voire impossible, d’avoir unedémarche exploratoire pour découvrir de nouveaux contenus. Les systèmes de recommandationsont des outils logiciels ayant pour objectif d’assister l’humain afin de répondre au problème desurcharge d’informations. Les travaux présentés dans ce document proposent une architecturepour la recommandation efficace d’articles d’actualité. L’approche ontologique utilisée repose surun modèle permettant une qualification précise des items sur la base d’un vocabulaire contrôlé.Contenu dans une ontologie, ce vocabulaire constitue une modélisation formelle de la vue métier surle domaine traité. Réalisés en collaboration avec la société Actualis SARL, ces travaux ont permis lacommercialisation d’un nouveau produit hautement compétitif, FristECO Pro’fil. / Effective management of large amounts of information has become a challenge increasinglyimportant for information systems. Everyday, new information sources emerge on the web. Someonecan easily find what he wants if (s)he seeks an article, a video or a specific artist. However,it becomes quite difficult, even impossible, to have an exploratory approach to discover newcontent. Recommender systems are software tools that aim to assist humans to deal withinformation overload. The work presented in this Phd thesis proposes an architecture for efficientrecommendation of news. In this document, we propose an architecture for efficient recommendationof news articles. Our ontological approach relies on a model for precise characterization of itemsbased on a controlled vocabulary. The ontology contains a formal vocabulary modeling a view on thedomain knowledge. Carried out in collaboration with the company Actualis SARL, this work has ledto the marketing of a new highly competitive product, FristECO Pro’fil.

[pt] Sistemas de recomendação aparecem em diversos domínios do mundo real. Vários modelos foram propostos para o problema de predição de entradas faltantes em um conjunto de dados. Duas das abordagens mais comuns são filtragem colaborativa baseada em similaridade e modelos de fatores latentes. Uma alternativa, mais recente, foi proposta por Salakhutdinov em 2007, usando máquinas de Boltzmann restritas, ou RBMs. Esse modelo se encaixa na família de modelos de fatores latentes, no qual, modelamos fatores latentes dos dados usando unidades binárias na camada escondida das RBMs. Esses modelos se mostraram capazes de aproximar resultados obtidos com modelos de fatoração de matrizes. Nesse trabalho vamos revisitar esse modelo e detalhar cuidadosamente como modelar e treinar RBMs para o problema de predição de entradas vazias em dados tabulares. / [en] Recommender systems can be used in many problems in the real world. Many models were proposed to solve the problem of predicting missing entries in a specific dataset. Two of the most common approaches are neighborhood-based collaborative filtering and latent factor models. A more recent alternative was proposed on 2007 by Salakhutdinov, using Restricted Boltzmann Machines. This models belongs to the family of latent factor models, in which, we model latent factors over the data using hidden binary units. RBMs have shown that they can approximate solutions trained with a traditional matrix factorization model. In this work we ll revisit this proposed model and carefully detail how to model and train RBMs for the problem of missing ratings prediction.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2017-07-12T13:47:04Z No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) risethesis.pdf: 1393384 bytes, checksum: f5f2fb9182ce60a9c5d2b0cd95f2893a (MD5) / Made available in DSpace on 2017-07-12T13:47:04Z (GMT). No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) risethesis.pdf: 1393384 bytes, checksum: f5f2fb9182ce60a9c5d2b0cd95f2893a (MD5) Previous issue date: 2016-06-29 / Em meio a grande sobrecarga de dados disponíveis na internet, sistemas de recomendação tornam-se ferramentas indispensáveis para auxiliar usuários no encontro de itens ou conteúdos relevantes. Diversas técnicas de recomendação são aplicadas em diversos tipos de domínios diferentes. Seja na recomendação de filmes, música, amigos, lugares ou notícias, sistemas de recomendação exploram diversas informações disponíveis para aprender as preferências dos usuários e promover recomendações úteis. Uma das estratégias mais utilizadas é a de filtragem colaborativa. A qualidade dessa estratégia depende da quantidade de avaliações disponíveis e da qualidade do algoritmo utilizado para predição de avaliação. Estudos recentes demonstram que informações provenientes de redes sociais podem ser muito úteis para aumentar a precisão das recomendações. Assim como acontece no mundo real, no mundo virtual usuários buscam recomendações e conselhos de amigos antes de comprar um item ou consumir algum serviço, informações desse tipo podem ser úteis para definição do contexto social da recomendação. Além do social, informações físicas e temporais passaram a ser utilizadas para definição do contexto físico de cada recomendação. A companhia, a localização e as condições climáticas são bons exemplos de elementos físicos que levam um usuário a preferir certos itens. Um processo de recomendação que não leve em consideração elementos contextuais pode fazer com que o usuário tenha uma péssima experiência consumindo determina do item recomendado equivocadamente. Esta dissertação tem como objetivo investigar técnicas de filtragem colaborativa que utilizam contexto a fim de realizar recomendações que auxiliem usuários no encontro de itens relevantes. Nesse tipo de técnica, um sistema de recomendação base é utilizando para fornecer recomendações para o usuário alvo. Em seguida, são filtrados apenas os itens considerados relevantes para contextos previamente identificados nas preferências do usuário alvo. As técnicas implementadas foram aplicadas em dois experimentos com duas bases de dados de domínios diferentes: uma base composta por eventos e outra por filmes. Na recomendação de eventos, investigamos o uso de contextos físicos (i.e., tempo e local) e de contextos sociais (i.e., amigos na rede social) associados aos itens sugeridos aos usuários. Na recomendação de filmes, por sua vez, investigamos novamente o uso de contexto social. A partir da aplicação de pós-filtragem em três algoritmos de filtragem colaborativa usados como base, foi possível recomendar itens de forma mais precisa, como demonstrado nos experimentos realizados. / The overload of data available on the internet makes recommendation systems become indispensable tools to assist users in meeting items or relevant content. Several recommendation techniques were has been userd in many different types of domains. Those systems can recommend movies, music, friends, places or news; recommender systems can exploit different information available to learn preferences of users and promote more useful recommendations. The collaborative filtering strategy is one of the most used. The quality of this technique depends on the number of available ratings and the algorithm used to predict. Recent studies show that information from social networks can be very useful to increase the accuracy recommendations. Just as in the real world, the virtual world users ask recommendations and advice from friends before buying an item or consume a service. Furthermore, the context of each rating may be crucial for the definition of new ratings. Location, date time and weather conditions are good examples of useful elements to define what should be the best items to recommend for some user. A recommendation process that does not respect those elements can provide a user a bad experience. This dissertation investigates collaborative filtering techniques based on context, and more specifically techniques based on post-filtering. First, a recommendation system was used to provide recommendations for a specific user. Then, only relevant items according to context preferences for the target user will be recommended. The techniques implemented was applied in two case studies with two different domains databases: one base composed of events and another of movies. In the event of recommendation, we investigated the use of physical contexts (i.e., time and place) and social contexts (i.e., friends in the social network) associated with items suggested to users. On the recommendation of movies, in turn, again we investigated the use of social context. From the application of post-filtering in three collaborative filtering algorithms used as a baseline, it was possible to recommend items more accurately, as demonstrated in the experiments.

Recommender systems are widely used in many domains. In this work, the importance of a recommender system in an online learning platform is discussed. After explaining the concept of adding an intelligent agent to online education systems, some features of the Course Networking (CN) website are demonstrated. Finally, the relation between CN, the intelligent agent (Rumi), and the recommender system is presented. Along with the argument of three different approaches for building a community recommendation system. The result shows that the Neighboring Collaborative Filtering (NCF) outperforms both the transfer learning method and the Continuous bag-of-words approach. The NCF algorithm has a general format with two various implementations that can be used for other recommendations, such as course, skill, major, and book recommendations.

Many employment services base their online job recommendations to users based solely on explicit data in their profiles. The implicit data of what users for example click on, save and mark as irrelevant goes unused. Instead of making recommendations based on user behavior they make a direct comparison between user preferences and job ad attributes. A reason for this is the concern that the inclusion of implicit data can give odd recommendations resulting in a loss of credibility for the service. However, as research has shown this to be of great advantage to recommender systems. In this paper I implement a job recommender and test it both with user data including interaction history with job ads as well as with only explicit data. The results of the recommender with implicit data got better overall performance, but negligible gain in the ratio between true and false positives, or in other words the ratio between correct and incorrect recommendations.