Biblio

Artificial neural networks are complex biologically inspired algorithms made up of highly distributed, adaptive and self-organizing structures that make them suitable for optimization problems. They are made up of a group of interconnected nodes, similar to the great networks of neurons in the human brain. So far, artificial neural networks have not been applied to user modeling in multi-criteria recommender systems. This paper presents neural networks-based user modeling technique that exploits some of the characteristics of biological neurons for improving the accuracy of multi-criteria recommendations. The study was based upon the aggregation function approach that computes the overall rating as a function of the criteria ratings. The proposed technique was evaluated using different evaluation metrics, and the empirical results of the experiments were compared with that of the single rating-based collaborative filtering and two other similarity-based modeling approaches. The two similarity-based techniques used are: the worst-case and the average similarity techniques. The results of the comparative analysis have shown that the proposed technique is more efficient than the two similarity-based techniques and the single rating collaborative filtering technique.

Recently, matrix factorization has produced state-of-the-art results in recommender systems. However, given the typical sparsity of ratings, the often large problem scale, and the large number of free parameters that are often implied, developing robust and efficient models remains a challenge. Previous works rely on dense and/or sparse factor matrices to estimate unavailable user ratings. In this work we develop a new formulation for recommender systems that is based on projective non-negative matrix factorization, but relaxes the non-negativity constraint. Driven by a simple yet instructive intuition, the proposed formulation delivers promising and stable results that depend on a minimal number of parameters. Experiments that we conducted on two popular recommender system datasets demonstrate the efficiency and promise of our proposed method. We make available our code and datasets at https://github.com/christosbampis/PCMF\_release.

Collaborative Filtering (CF) is a successful technique that has been implemented in recommender systems and Privacy Preserving Collaborative Filtering (PPCF) aroused increasing concerns of the society. Current solutions mainly focus on cryptographic methods, obfuscation methods, perturbation methods and differential privacy methods. But these methods have some shortcomings, such as unnecessary computational cost, lower data quality and hard to calibrate the magnitude of noise. This paper proposes a (k, p, I)-anonymity method that improves the existing k-anonymity method in PPCF. The method works as follows: First, it applies Latent Factor Model (LFM) to reduce matrix sparsity. Then it improves Maximum Distance to Average Vector (MDAV) microaggregation algorithm based on importance partitioning to increase homogeneity among records in each group which can retain better data quality and (p, I)-diversity model where p is attacker's prior knowledge about users' ratings and I is the diversity among users in each group to improve the level of privacy preserving. Theoretical and experimental analyses show that our approach ensures a higher level of privacy preserving based on lower information loss.

Much of the focus of recommender systems research has been on the accurate prediction of users' ratings for unseen items. Recent work has suggested that objectives such as diversity and novelty in recommendations are also important factors in the effectiveness of a recommender system. However, methods that attempt to increase diversity of recommendation lists for all users without considering each user's preference or tolerance for diversity may lead to monotony for some users and to poor recommendations for others. Our goal in this research is to evaluate the hypothesis that users' propensity towards diversity varies greatly and that the diversity of recommendation lists should be consistent with the level of user interest in diverse recommendations. We propose a pre-filtering clustering approach to group users with similar levels of tolerance for diversity. Our contributions are twofold. First, we propose a method for personalizing diversity by performing collaborative filtering independently on different segments of users based on the degree of diversity in their profiles. Secondly, we investigate the accuracy-diversity tradeoffs using the proposed method across different user segments. As part of this evaluation we propose new metrics, adapted from information retrieval, that help us measure the effectiveness of our approach in personalizing diversity. Our experimental evaluation is based on two different datasets: MovieLens movie ratings, and Yelp restaurant reviews.

In recommender systems based on low-rank factorization of a partially observed user-item matrix, a common phenomenon that plagues many otherwise effective models is the interleaving of good and spurious recommendations in the top-K results. A single spurious recommendation can dramatically impact the perceived quality of a recommender system. Spurious recommendations do not result in serendipitous discoveries but rather cognitive dissonance. In this work, we investigate folding, a major contributing factor to spurious recommendations. Folding refers to the unintentional overlap of disparate groups of users and items in the low-rank embedding vector space, induced by improper handling of missing data. We formally define a metric that quantifies the severity of folding in a trained system, to assist in diagnosing its potential to make inappropriate recommendations. The folding metric complements existing information retrieval metrics that focus on the number of good recommendations and their ranks but ignore the impact of undesired recommendations. We motivate the folding metric definition on synthetic data and evaluate its effectiveness on both synthetic and real world datasets. In studying the relationship between the folding metric and other characteristics of recommender systems, we observe that optimizing for goodness metrics can lead to high folding and thus more spurious recommendations.

Trust networks have been widely used to mitigate the data sparsity and cold-start problems of collaborative filtering. Recently, some approaches have been proposed which exploit explicit signed trust relationships, i.e., trust and distrust relationships. These approaches ignore the fact that users despite trusting/distrusting each other in a trust network may have different preferences in real-life. Most of these approaches also handle the notion of the transitivity of distrust as well as trust. However, other existing work observed that trust is transitive while distrust is intransitive. Moreover, explicit signed trust relationships are fairly sparse and may not contribute to infer true preferences of users. In this paper, we propose to create implicit signed trust relationships and exploit them along with explicit signed trust relationship to solve sparsity problem of trust relationships. We also confirm the similarity (resp. dissimilarity) of implicit and explicit trust (resp. distrust) relationships by using the similarity score between users so that users' true preferences can be inferred. In addition to these strategies, we also propose a matrix factorization model that simultaneously exploits implicit and explicit signed trust relationships along with rating information and also handles transitivity of trust and intransitivity of distrust. Extensive experiments on Epinions dataset show that the proposed approach outperforms existing approaches in terms of accuracy.

Recommender system is to suggest items that might be interest of the users in social networks. Collaborative filtering is an approach that works based on similarity and recommends items liked by other similar users. Trust model adopts users' trust network in place of similarity. Multi-faceted trust model considers multiple and heterogeneous trust relationship among the users and recommend items based on rating exist in the network of trustees of a specific facet. This paper applies genetic algorithm to estimate parameters of multi-faceted trust model, in which the trust weights are calculated based on the ratings and the trust network for each facet, separately. The model was built on Epinions data set that includes consumers' opinion, rating for items and the web of trust network. It was used to predict users' rating for items in different facets and root mean squared of prediction error (RMSE) was considered as a measure of performance. Empirical evaluations demonstrated that multi-facet models improve performance of the recommender system.

The rapid development of cloud computing has resulted in the emergence of numerous web services on the Internet. Selecting a suitable cloud service is becoming a major problem for users especially non-professionals. Quality of Service (QoS) is considered to be the criterion for judging web services. There are several Collaborative Filtering (CF)-based QoS prediction methods proposed in recent years. QoS values among different users may vary largely due to the network and geographical location. Moreover, QoS data provided by untrusted users will definitely affect the prediction accuracy. However, most existing methods seldom take both facts into consideration. In this paper, we present a trust-aware and location-based approach for web service QoS prediction. A trust value for each user is evaluated before the similarity calculation and the location is taken into account in similar neighbors selecting. A series of experiments are performed based on a realworld QoS dataset including 339 service users and 5,825 services. The experimental analysis shows that the accuracy of our method is much higher than other CF-based methods.

Route planner systems support commuters and city visitors in finding the best route between two arbitrary points. More advanced route planners integrate different transportation modes such as private transport, public transport, car- and bicycle sharing or walking and are able combine these to multi-modal routes. Nevertheless, state-of-the-art planner systems usually do not consider the users' personal preferences or the wisdom of the crowd when suggesting multi-modal routes. Including the knowledge and experience of locals who are familiar with local transport allows identification of alternative routes which are, for example, less crowded during peak hours. Collaborative filtering (CF) is a technique that allows recommending items such as multi-modal routes based on the ratings of users with similar preferences. In this paper, we introduce RouteMe, a mobile recommender system for personalized, multi-modal routes which combines CF with knowledge-based recommendations to increase the quality of route recommendations. We present our hybrid algorithm in detail and show how we integrate it in a working prototype. The results of a user study show that our prototype combining CF, knowledge-based and popular route recommendations outperforms state-of-the-art route planners.

Recommender systems are typically evaluated on their ability to provide items that satisfy the needs and interests of the end user. However, in many real world applications, users are not the only stakeholders involved. There may be a variety of individuals or organizations that benefit in different ways from the delivery of recommendations. In this paper, we re-define the recommender system as a multistakeholder environment in which different stakeholders are served by delivering recommendations, and we suggest a utility-based approach to evaluating recommendations in such an environment that is capable of distinguishing among the distributions of utility delivered to different stakeholders.

Due to the large quantity and diversity of content being easily available to users, recommender systems (RS) have become an integral part of nearly every online system. They allow users to resolve the information overload problem by proactively generating high-quality personalized recommendations. Trust metrics help leverage preferences of similar users and have led to improved predictive accuracy which is why they have become an important consideration in the design of RSs. We argue that there are additional aspects of trust as a human notion, that can be integrated with collaborative filtering techniques to suggest to users items that they might like. In this paper, we present an approach for the top-N recommendation task that computes prediction scores for items as a user specific combination of global and local trust models to capture differences in preferences. Our experiments show that the proposed method improves upon the standard trust model and outperforms competing top-N recommendation approaches on real world data by upto 19%.

Application of trust principals in internet of things (IoT) has allowed to provide more trustworthy services among the corresponding stakeholders. The most common method of assessing trust in IoT applications is to estimate trust level of the end entities (entity-centric) relative to the trustor. In these systems, trust level of the data is assumed to be the same as the trust level of the data source. However, most of the IoT based systems are data centric and operate in dynamic environments, which need immediate actions without waiting for a trust report from end entities. We address this challenge by extending our previous proposals on trust establishment for entities based on their reputation, experience and knowledge, to trust estimation of data items [1-3]. First, we present a hybrid trust framework for evaluating both data trust and entity trust, which will be enhanced as a standardization for future data driven society. The modules including data trust metric extraction, data trust aggregation, evaluation and prediction are elaborated inside the proposed framework. Finally, a possible design model is described to implement the proposed ideas.

Collaborative filtering plays an essential role in a recommender system, which recommends a list of items to a user by learning behavior patterns from user rating matrix. However, if an attacker has some auxiliary knowledge about a user purchase history, he/she can infer more information about this user. This brings great threats to user privacy. Some methods adopt differential privacy algorithms in collaborative filtering by adding noises to a rating matrix. Although they provide theoretically private results, the influence on recommendation accuracy are not discussed. In this paper, we solve the privacy problem in recommender system in a different way by applying the differential privacy method into the procedure of recommendation. We design two differentially private recommender algorithms with sampling, named Differentially Private Item Based Recommendation with sampling (DP-IR for short) and Differentially Private User Based Recommendation with sampling(DP-UR for short). Both algorithms are based on the exponential mechanism with a carefully designed quality function. Theoretical analyses on privacy of these algorithms are presented. We also investigate the accuracy of the proposed method and give theoretical results. Experiments are performed on real datasets to verify our methods.

We provide a generic framework that, with the help of a preprocessing phase that is independent of the inputs of the users, allows an arbitrary number of users to securely outsource a computation to two non-colluding external servers. Our approach is shown to be provably secure in an adversarial model where one of the servers may arbitrarily deviate from the protocol specification, as well as employ an arbitrary number of dummy users. We use these techniques to implement a secure recommender system based on collaborative filtering that becomes more secure, and significantly more efficient than previously known implementations of such systems, when the preprocessing efforts are excluded. We suggest different alternatives for preprocessing, and discuss their merits and demerits.