Submitted by Erika Demachki (erikademachki@gmail.com) on 2015-02-06T20:35:15Z No. of bitstreams: 2 Dissertação - Everton Lima Aleixo - 2014.pdf: 2375638 bytes, checksum: accbd56745e040e23362d951a1336538 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Erika Demachki (erikademachki@gmail.com) on 2015-02-06T20:35:41Z (GMT) No. of bitstreams: 2 Dissertação - Everton Lima Aleixo - 2014.pdf: 2375638 bytes, checksum: accbd56745e040e23362d951a1336538 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-02-06T20:35:41Z (GMT). No. of bitstreams: 2 Dissertação - Everton Lima Aleixo - 2014.pdf: 2375638 bytes, checksum: accbd56745e040e23362d951a1336538 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2014-09-02 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Memory-based algorithms are the most popular among the collaborative filtering algorithms. They use as input a table containing ratings given by users to items, known as the rating matrix. They predict the rating given by user a to an item i by computing similarities of the ratings among users or similarities of the ratings among items. In the first case Memory-Based algorithms are classified as User-based algorithms and in the second one they are labeled as Item-based algorithms. The prediction is computed using the ratings of k most similar users (or items), also know as neighbors. Memory-based algorithms are simple to understand and to program, usually provide accurate recommendation and are less sensible to data change. However, to obtain the most similar neighbors for a prediction they have to process all the data which is a serious scalability problem. Also they are sensitive to the sparsity of the input. In this work we propose an efficient and effective Item-Based that aims at diminishing the sensibility of the Memory-Based approach to both problems stated above. The algorithm is faster (almost 50%) than the traditional Item-Based algorithm while maintaining the same level of accuracy. However, in environments that have much data to predict and few to train the algorithm, the accuracy of the proposed algorithm surpass significantly that of the traditional Item-based algorithms. Our approach can also be easily adapted to be used as User-based algorithms. / Algoritmos baseados em memória são os mais populares entre os algoritmos de filtragem colaborativa. Eles usam como entrada uma tabela contendo as avaliações feitas pelos usuários aos itens, conhecida como matriz de avaliações. Eles predizem a avaliação dada por um usuário a a um item i, computando a similaridade de avaliações entre a e outros usuários ou entre i e outros itens. No primeiro caso, os algoritmos baseados em memória são classificados como algoritmos baseados em usuários (User-based) e no segundo caso são rotulados como algoritmos baseados em itens (Item-Based). A predição é computada usando as avaliações dos k usuários (ou itens) mais similares, também conhecidos como vizinhos. Algoritmos baseados em memória são simples de entender e implementar. Normalmente produzem boas recomendações e são menos sensíveis a mudança nos dados. Entretanto, para obter os vizinhos mais similares para a predição, eles necessitam processar todos os dados da matriz, o que é um sério problema de escalabilidade. Eles também são sensíveis a densidade dos dados. Neste trabalho, nós propomos um algoritmo eficiente e eficaz baseado em itens que visa diminuir a sensibilidade dos algoritmos baseados em memória para ambos os problemas acima referidos. Esse algoritmo é mais rápido (quase 50%) do que o algoritmo baseado em itens tradicional, mantendo o mesmo nível de acurácia. Entretanto, em ambientes onde existem muitos dados para predizer e poucos para treinar o algoritmo, a acurácia do algoritmo proposto supera significativamente a do algoritmo tradicional baseado em itens. Nossa abordagem pode ainda ser facilmente adaptada para ser utilizada como o algoritmo baseado em usuários.

abstract: Social media has become the norm of everyone for communication. The usage of social media has increased exponentially in the last decade. The myriads of Social media services such as Facebook, Twitter, Snapchat, and Instagram etc allow people to connect with their friends, and followers freely. The attackers who try to take advantage of this situation has also increased at an exponential rate. Every social media service has its own recommender systems and user profiling algorithms. These algorithms use users current information to make different recommendations. Often the data that is formed from social media services is Linked data as each item/user is usually linked with other users/items. Recommender systems due to their ubiquitous and prominent nature are prone to several forms of attacks. One of the major form of attacks is poisoning the training set data. As recommender systems use current user/item information as the training set to make recommendations, the attacker tries to modify the training set in such a way that the recommender system would benefit the attacker or give incorrect recommendations and hence failing in its basic functionality. Most existing training set attack algorithms work with ``flat" attribute-value data which is typically assumed to be independent and identically distributed (i.i.d.). However, the i.i.d. assumption does not hold for social media data since it is inherently linked as described above. Usage of user-similarity with Graph Regularizer in morphing the training data produces best results to attacker. This thesis proves the same by demonstrating with experiments on Collaborative Filtering with multiple datasets. / Dissertation/Thesis / Masters Thesis Computer Science 2019

Privacy is defined as the freedom from unauthorized intrusion. The availability of personal information through online databases, such as government records, medical records, and voters and #146; lists, pose a threat to personal privacy. The concern over individual privacy has led to the development of legal codes for safeguarding privacy in several countries. However, the ignorance of individuals as well as loopholes in the systems, have led to information breaches even in the presence of such rules and regulations. Protection against data privacy requires modification of the data itself. The term {em data obfuscation} is used to refer to the class of algorithms that modify the values of the data items without distorting the usefulness of the data. The main goal of this thesis is the development of a data obfuscation technique that provides robust privacy protection with minimal loss in usability of the data. Although medical and financial services are two of the major areas where information privacy is a concern, privacy breaches are not restricted to these domains. One of the areas where the concern over data privacy is of growing interest is collaborative filtering. Collaborative filtering systems are being widely used in E-commerce applications to provide recommendations to users regarding products that might be of interest to them. The prediction accuracy of these systems is dependent on the size and accuracy of the data provided by users. However, the lack of sufficient guidelines governing the use and distribution of user data raises concerns over individual privacy. Users often provide the minimal information that is required for accessing these E-commerce services. The lack of rules governing the use and distribution of data disallows sharing of data among different communities for collaborative filtering. The goals of this thesis are (a) the definition of a standard for classifying DO techniques, (b) the development of a robust cluster preserving data obfuscation algorithm, and (c) the design and implementation of a privacy-preserving shared collaborative filtering framework using the data obfuscation algorithm.

As the Internet became widespread all over the world, people started to share great amount of data on the web and almost every people joined different data networks in order to have a quick access to data shared among people and survive against the information overload on the web. Recommender systems are created to provide users more personalized information services and to make data available for people without an extra effort. Most of these systems aim to get or learn user preferences, explicitly or implicitly depending to the system, and guess &ldquo / preferable data&rdquo / that has not already been consumed by the user. Traditional approaches use user/item similarity or item content information to filter items for the active user / however most of the recent approaches also consider the trustworthiness of users. By using trustworthiness, only reliable users according to the target user opinion will be considered during information retrieval. Within this thesis work, a content boosted method of using trust data in recommender systems is proposed. It is aimed to be shown that people who trust the active user and the people, whom the active user trusts, also have correlated opinions with the active user. This results the fact that the rated items by these people can also be used while offering new items to users. For this research, www.epinions.com site is crawled, in order to access user trust relationships, product content information and review ratings which are ratings given by users to product reviews that are written by other users.

Data analysis has become a very important area for both companies and researchers as a consequence of the technological developments in recent years. Companies are trying to increase their profit by analyzing the existing data about their customers and making decisions for the future according to the results of these analyses. Parallel to the need of companies, researchers are investigating different methodologies to analyze data more accurately with high performance. Recommender systems are one of the most popular and widespread data analysis tools. A recommender system applies knowledge discovery techniques to the existing data and makes personalized product recommendations during live customer interaction. However, the huge growth of customers and products especially on the internet, poses some challenges for recommender systems, producing high quality recommendations and performing millions of recommendations per second. In order to improve the performance of recommender systems, researchers have proposed many different methods. Singular Value Decomposition (SVD) technique based on dimension reduction is one of these methods which produces high quality recommendations, but has to undergo very expensive matrix calculations. In this thesis, we propose and experimentally validate some contributions to SVD technique which are based on the user and the item categorization. Besides, we adopt tags to classical 2D (User-Item) SVD technique and report the results of experiments. Results are promising to make more accurate and scalable recommender systems.

Searching the Web involves more than sifting through a huge graph of pages and hyperlinks. Specific collaboration networks have emerged that serve domain-specific queries better by exploiting the principles and patterns that apply there. We continue this trend by suggesting heuristics and algorithms to mine the evolution of collaboration networks, to discover interesting roles played by entities. The first section of the dissertation introduces the concept of nurturers using the computer science research community as a case study, while the second section formulates three roles - scouts, promoters and connectors, played by ratings in collaborative filtering systems. Nurturers: Nurturing, a pervasive mammalian trait, naturally extends to most association networks that involve humans. The increased availability of digital and online data about associations lets researchers experiment with algorithms to gain insight into such phenomena. Consider some examples of nurturing: • Slashdot endorsement. Slashdot was not the first site to link to Firefox, but the publicity Firefox received from this association surely helped it become popular quickly. The phenomenon of many small websites crashing due to publicity received through Slashdot has become well known as the Slashdot Effect. • A VC (Venture Capitalist) seed-funding a new startup. This event has a high nurturing value if the startup’s valuation increases rapidly after the funding. • A blogger writing about a topic. Kim Cameron has nurtured the “Laws of Identity” topic if it later becomes the buzz in blog circles. A nurturer need not always be the innovator or originator. The evangelist who adopts a prodigal idea and launches it on its way to success can also be a nurturer. • A professor guiding his student through the art of scientific research and bootstrapping him into a vibrant research community. New nodes not only emerge around these nurturers, but also become important in the network. Knowing nurturers is useful especially in vertical search, where algorithms exploit the structure of specialized collaboration networks to make search more relevant: knowing early adopters of good web pages can make web-search fresher; a list of VCs ranked by their nurturing value is useful to people with new startup ideas; the list of top nurturers in computer science is a valuable resource for a student seeking to do research. This dissertation presents a framework for discovering nurturers by mining the evolution of an association network, and discusses heuristics and customizations that can be applied through a case study: finding the Best Nurturers in Computer Science Research. Roles of Ratings in Collaborative Filtering: Recommender systems aggregate individual user ratings into predictions of products or services that might interest visitors. The quality of this aggregation process crucially affects user experience and hence the effectiveness of recommenders in e-commerce. The dissertation presents a novel study that disaggregates global recommender performance metrics into contributions made by each individual rating, allowing us to characterize the many roles played by ratings in nearest neighbor collaborative filtering. In particular, we formulate three roles - scouts, promoters, and connectors that capture how users receive recommendations, how items get recommended, and how ratings of these two types are themselves connected (respectively). These roles find direct uses in improving recommendations for users, in better targeting of items, and most impor -tantly, in helping monitor the health of the system as a whole. For instance, they can be used to track the evolution of neighborhoods, to identify rating subspaces that do not contribute (or contribute negatively) to system performance, to enumerate users who are in danger of leaving, and to assess the susceptibility of the System to attacks such as shilling. The three rating roles presented here provide broad primitives to manage a recommender system and its community.

Recommendation systems are emerging as an important business application as the demand for personalized services in E-commerce increases. Collaborative filtering techniques are widely used for predicting a user's preference or generating a list of items to be recommended. In this thesis, we develop several new approaches for collaborative filtering based on model combination and kernel smoothing. Specifically, we start with an experimental study that compares a wide variety of CF methods under different conditions. Based on this study, we formulate a combination model similar to boosting but where the combination coefficients are functions rather than constant. In another contribution we formulate and analyze a local variation of matrix factorization. This formulation constructs multiple local matrix factorization models and then combines them into a global model. This formulation is based on the local low-rank assumption, a slightly different but more plausible assumption about the rating matrix. We apply this assumption to both rating prediction and ranking problems, with both empirical validations and theoretical analysis. We contribute with this thesis in four aspects. First, the local approaches we present significantly improve the accuracy of recommendations both in rating prediction and ranking problems. Second, with the more realistic local low-rank assumption, we fundamentally change the underlying assumption for matrix factorization-based recommendation systems. Third, we present highly efficient and scalable algorithms which take advantage of parallelism, suited for recent large scale datasets. Lastly, we provide an open source software implementing the local approaches in this thesis as well as many other recent recommendation algorithms, which can be used both in research and production.