Abstract. Implicit acquisition of user preferences makes log-based collaborative filtering favorable in practice to accomplish recommendations. In this paper, we follow a formal approach in text retrieval to re-formulate the problem. Based on the classic probability ranking principle, we propose a probabilistic user-item relevance model. Under this formal model, we show that user-based and item-based approaches are only two different factorizations with different independence assumptions. Moreover, we show that smoothing is an important aspect to estimate the parameters of the models due to data sparsity. By adding linear interpolation smoothing, the proposed model gives a probabilistic justification of using TF×IDF-like item ranking in collaborative filtering. Besides giving the insight understanding of the problem of collaborative filtering, we also show experiments in which the proposed method provides a better recommendation performance on a music play-list data set. 1

We consider in this paper a popular class of recommender systems that are based on Collaborative Filtering (CF for short). CF is the process of predicting customer ratings to items based on previous ratings of (similar) users to (similar) items, and is typically used by a single organization, using its own customer ratings. We argue here that a multi-organization collaboration, even for organizations operating in different subject domains, can greatly improve the quality of the recommendations that the individual organizations provide to their users. To substantiate this claim, we present C2F (Collaborative CF), a recommender system that retains the simplicity and efficiency of classical CF, while allowing distinct organizations to collaborate and boost their recommendations. C2F employs CF in a distributed fashion that improves the quality of the generated recommendations, while minimizing the amount of data exchanged between the collaborating parties. Key ingredient of the solution are succinct signatures that can be computed locally for items (users) in a given organization and suffice for identifying similar items (users) in the collaborating organizations. We show that the use of such compact signatures not only reduces data exchange but also allows to speed up, by over 50%, the recommendations computation time. 1.

[Towards truly personalized TV] Intensive research efforts in the field of multimedia content analysis in the past 15 years have resulted in an abundance of theoretical and algorithmic solutions for extracting the content-related information from audiovisual signals. The solutions proposed so far cover an enormous application scope and aim at enabling us to easily access the events, people, objects, and scenes captured by the camera, to quickly retrieve our favorite themes from a large music video archive (e.g., a pop/rock concert database), or to efficiently generate comprehensive overviews, summaries, and abstracts of movies, sports TV broadcasts, surveillance, meeting recordings, and educational video material. However, what about the task of finding exciting parts of a sports TV broadcast or funny and romantic excerpts from a movie? What about locating unpleasant video clips we would be reluctant to let our children watch? This article considers how we feel about the content we see or hear. As opposed to the cognitive content information composed of the facts about the genre, temporal content structure (shots, scenes) and spatiotemporal content elements (objects, persons, events, topics) we are interested in obtaining the information about the feelings, emotions, and moods evoked by a speech, audio, or video clip. We refer to the latter as the affective content, and to the terms such as “happy ” or “exciting ” as the affective labels of an audiovisual signal.

The decentralized architecture of peer-to-peer (P2P) networks solves many of the limitations of conventional client-server networks. This decentralization, however, creates the need in P2P file sharing networks to find peers who are downloading the same file, a problem which is referred to as swarm discovery. In the BitTorrent file sharing network, swarm discovery is solved using a central server (a tracker), which is unreliable and unscalable. We have designed a decentralized swarm discovery protocol, called LITTLE BIRD. LITTLE BIRD is an epidemic protocol that exchanges swarm information between peers that are currently or were recently members of a swarm. A quantitative measure of the contribution of each peer is calculated to make the protocol efficient and secure against peers that pollute the system with malicious information. Furthermore, we have conducted detailed measurements of the BitTorrent community ’Filelist.org’, in order to study download swarm behavior and optimize the design of our protocol. We have implemented the LITTLE BIRD protocol as an addition to the Tribler P2P network.

Noname manuscript No. (will be inserted by the editor) A unifying framework of rating users and data items in

Abstract — In Peer-to-Peer (P2P) file sharing systems, peers spend a significant amount of time looking for relevant and interesting files. However, the files available for download represent on one hand a rich collection for different needs and preferences and on the other hand a struggle for the peers to find files that they like. In this paper, we propose new recommender schemes based on collaborative filtering. Peers collaborate to filter out irrelevant files. These schemes help peers find and discover new, interesting and relevant files. We propose recommender schemes based on Files ’ Popularity and/or Peers ’ Similarity. To overcome the problems of traditional collaborative filtering recommender systems, an implicit rating approach is used. Simulation results confirm the effectiveness of Peers ’ Similarity based Recommendation in providing accurate recommendations. In addition, the proposed recommender schemes are proactive. Recommendations are provided to peers to motivate them to download the recommended files. I.

Recommendation services are increasingly becoming indispensable part of people’s everyday online experiences. However, serious issues on customer data privacy and ownership, recommendation security, and difficulty in generating cross-domain and cross-site recommendations arise with the centralized server-based architecture currently employed by online service and product providers. We propose a distributed recommender system based on the construction and maintenance of a user similarity network in which each user only maintains a small number of close neighbors for peer-to-peer neighbor discovery and recommendation generation. Empirical results using the Netflix dataset showed that our proposed system achieved comparable recommendation quality as the centralized recommendation with significantly fewer similarity calculations to identify user neighbor sets for recommendation generation. 1.

This thesis examines reasoning under uncertainty in distributed systems. Unlike in centralized systems, where the observations reside in a single location, the observations in distributed systems are often scattered across the network. To reason accurately, a networked device often needs to incorporate observations from other nodes and must do so with limited computation and communication even for large problems. The reasoning is further complicated by unstable network conditions, characteristic to many real-world networks: the nodes may fail, communication links may become unreliable, and the entire network may get fragmented into several components that cannot communicate with each other. These aspects make distributed inference very challenging. We consider one general problem of distributed filtering for estimating the state of a dynamical system and three independent applications: simultaneous localization and tracking, where a camera network localizes itself by observing a moving object, internal localization of large-scale modular robots, where a robot determines the relative poses of its internal parts,