Search

Recommender systems are proving to be an invaluable tool for extracting user-relevant content helping users in their daily activities (e.g., finding relevant places to visit, content to consume, items to purchase). However, to be effective, these systems need to collect and analyze large volumes of personal data (e.g., location check-ins, movie ratings, click rates .. etc.), which exposes users to numerous privacy threats. In this context, recommender systems based on Federated Learning (FL) appear to be a promising solution for enforcing privacy as they compute accurate recommendations while keeping personal data on the users' devices. However, FL, and therefore FL-based recommender systems, rely on a central server that can experience scalability issues besides being vulnerable to attacks. To remedy this, we propose PEPPER, a decentralized recommender system based on gossip learning principles. In PEPPER, users gossip model updates and aggregate them asynchronously. At the heart of PEPPER reside two key components: a personalized peer-sampling protocol that keeps in the neighborhood of each node, a proportion of nodes that have similar interests to the former and a simple yet effective model aggregation function that builds a model that is better suited to each user. Through experiments on three real datasets implementing two use cases: a location check-in recommendation and a movie recommendation, we demonstrate that our solution converges up to 42% faster than with other decentralized solutions providing up to 9% improvement on average performance metric such as hit ratio and up to 21% improvement on long tail performance compared to decentralized competitors., Comment: 23 pages, 15 figures

The widespread adoption of handheld devices (e.g., smartphones, tablets) makes mobility traces of users broadly available to third party services. These traces are collected by means of various sensors embedded in the users’ devices, including GPS, WiFi and GSM. We study in this paper the mobility of 300 users over a period up to 31 months from the perspective of the above three types of data and with a focus on two cities, i.e., Lausanne (Switzerland) and Lyon (France). We found that users’ mobility traces, no matter if they are collected using GPS, WiFi or GSM antennas, are highly unique. We show that on average only four spatio-temporal points from the WiFi, GSM and GPS traces are enough to uniquely identify 94% of the individuals, on both datasets. In addition, we show that using the temporal dimension (i.e., whether users move or are in a meaningful location such as their home or their working place) drastically improves the capacity to uniquely identify them compared to when only exploiting the spatial dimension (by 14% on average). In some cases, using the temporal dimension alone can represent a better mobility footprint than the spatial dimension to discriminate users. We further conduct a de-anonymisation attack to assess how mobility traces can be re-identified, and show that almost all users can be de-anonymised with a high success rate. Finally, we apply different Location Privacy Protection Mechanisms (LPPMs), including spatial filtering, temporal cloaking, adding spatial noise to mobility data, or using generalisation, and analyse the impact of these mechanisms on both the uniqueness of users’ mobility traces and the outcome of the de-anonymisation attack. We show that spatially obfuscating mobility data is not enough to protect users, and that classical LPPMs are not able to protect users against a de-anonymisation attack. We finally conclude this paper by drawing some insights towards future spatio-temporal LPPMs.

International audience; Video on Demand (VoD) streaming is the largest source of Inter-net traffic. Efficient and scalable VoD requires Content Delivery Networks (CDNs) whose cost are prohibitive for many providers. An alternative is to cache and serve video content using end-users devices. Direct connections between these devices complement the resources of core VoD servers with an edge-assisted collaborative CDN. VoD access histories can reveal critical personal information, and centralized VoD solutions are notorious for exploiting personal data. Hiding the interests of users from servers and edge-assisting devices is necessary for a new generation of privacy-preserving VoD services. We introduce PrivaTube, a scalable and cost-effective VoD solution. PrivaTube aggregates video content from multiple servers and edge peers to offer a high Quality of Experience (QoE) for its users. It enables privacy preservation at all levels of the content distribution process. It leverages Trusted Execution Environments (TEEs) at servers and clients, and obfuscates access patterns using fake requests that reduce the risk of personal information leaks. Fake requests are further leveraged to implement proactive provisioning and improve QoE. Our evaluation of a complete prototype shows that PrivaTube reduces the load on servers and increases QoE while providing strong privacy guarantees.

International audience; Since the advent of hand held devices (e.g., smartphones, tablets, smart watches) with Ubiquitous computing and the wide popularity of location-based mobile applications, the amount of captured user location data is dramatically increasing. However, the gathering and exploitation of this data by mobile application providers raises many privacy threats as sensitive information can be inferred from it (e.g., home and work locations, religious beliefs, sexual orienta-tions and social relationships). To address this issue a number of data obfuscation techniques (also called Location Privacy Protection Mechanisms or LPPMs) have been proposed in the literature. One of the existing methods to assess the effectiveness of LPPMs is to test them against user re-identification attacks. The aim of these attacks is to break user anonymity by re-associating data obfuscated using a given LPPM with user profiles built from user past mobility. In this paper, we present AP-Attack a novel re-identification attack that relies on a heatmap representation of user mobility data. Our experiments run against three representative LPPMs of the literature using four real mobility datasets show that AP-Attack succeeds in re-identifying up to 79% users in non-obfuscated data, +27% more users than POI-Attack and PIT-Attack two well known state-of-the-art attacks. We also present a simple technique to improve user protection against our attack, which relies on a user-centric application of multiple-LPPMs.

International audience; Leisure activities are essential in the individual's pursuit of well-being and higher satisfaction with life. In the last few years, the event-based social networks and event-promoting services on Face-book, Couchsurfing etc., have seen a boosting use. Their users are looking for personalised experience but they are often overwhelmed with the number of available options. To overcome this issue, rec-ommender systems serve as users' personal guides. They exploit various techniques and types of influence (geographic, temporal, social, etc.). In this work, we investigate the impact of individu-als' psychological profiles on their selection of leisure activities. We describe a comprehensive user study that we have conducted and an associated collection that could be further exploited for the recommendation of leisure activities.

The main objective of this research was to identify co-referent entities located in several linked open data (LOD) sources that are described in various natural languages. The problem is approached from two perspectives. First, we do a multi-scale analysis of the RDF graph to discover structural similarities of entities. This was implemented as a tensor decomposition of the RDF graph with each predicate corresponding to a horizontal slide of the tensor. Hereafter, we used the term "structural evidence" to refer to the result of this analysis. Second, for each entity, we associated textual data coming from the Web of documents. Thus, after some preprocessing (viz. removing empty words, applying a weighting scheme such as tf-idf, ...), we represented each entity in a high dimensional space with each dimension corresponding to a term. Next, through a Singular Value Decomposition (SVD), we find a subspace such that the sum of squared distances from the original space to the sub space is minimized. This dimensionality reduction allows us to find language independent similarities between entities. Hereafter, we use the term "textual evidence" to refer to the result of this analysis. Since the similarity information coming from the structural and the textual evidence are complementary to each other, a global similarity score is computed by aggregating the two evidences. We adopt a linear opinion pool, an approach which is commonly used in belief aggregation as an aggregation mechanism. In the end, for any given entity, we obtained a global similarity vector. The higher component values of this vector correspond to potential co-referent entities. All algorithms are implemented in Python. According to the experiment result conducted on the French and English DBpedia, our approach can bring high results.

International audience; In this paper, a dataset of geotagged photos on a world-wide scale is presented. The dataset contains a sample of more than 14 Million geotagged photos crawled from Flickr with the corresponding metadata. To guarantee the spatial representativeness of the dataset, a crawling approach based on the small-world phenomena and the Flickr friendship's graph is applied. Furthermore, the noisiness of user-provided tags is reduced through an automatic tag cleaning approach. To enable efficient retrieval, photos in the dataset are indexed based on their location information using quad-tree data structure. The dataset can assists different applications, especially, search-based automatic image annotation and reverse geotagging.

International audience; This paper describes a mobile Web application that allows browsing conference publications, their authors, authors’ organizations, and even authors’ other publications or publications related to the same keywords. It queries a main SPARQL endpoint that serves the conference metadata set, as well as other endpoints to enrich and explore data. It provides extra functions, such as flashing a publication QR code from the Web browser, accessing external resources about the publications, and it can be linked to external Web services. This application exploits the Linked Data paradigm and performs client-side reasoning. It follows recent W3C technical advances and as a mashup, requires few server resources. It can easily be deployed for any conference with available metadata on the Web.

International audience; In this article, we propose an evolution of trust-based rec- ommender systems that only relies on local information and can be deployed on top of existing social networks. Our ap- proach takes into account friends' similarity and confidence on ratings, but limits data exchange to direct friends, in order to prevent ratings from being globally known. There- fore, calculations are limited to locally processed algorithms, privacy concerns can be taken into account and algorithms are suitable for decentralized or peer-to-peer architectures. We have implemented and evaluated our approach against five others, using the Epinions trust network. We show that local information with good default scoring strategies are sufficient to cover more users than classical collaborative filtering and trust-based recommender systems. Regarding accuracy, our approach performs better than most others, specially for cold start users, despite using less information.

International audience; A huge number of user-tagged images are daily uploaded to the web. Recently, a growing number of those images are also geotagged. These provide new opportunities for solutions to automatically tag images so that efficient image management and retrieval can be achieved. In this paper an automatic image annotation approach is proposed. It is based on a statistical model that combines two different kinds of information: high level information represented by user tags of images captured in the same location as a new unlabeled image (input image); and low level information represented by the visual similarity between the input image and the collection of geographically similar images. To maximize the number of images that are visually similar to the input image, an iterative visual matching approach is proposed and evaluated. The results show that a significant recall improvement can be achieved with an increasing number of iterations. The quality of the recommended tags has also been evaluated and an overall good performance has been observed.

International audience; The vision of Universal Multimedia Access (UMA) and Universal Multimedia Experience (UME) has driven research in the multimedia community for a long time. Implementing content adaptation frameworks to satisfy heterogeneous types of constraints is among the main requirements for UMA and UME. At the core of these frameworks lies the adaptation decision engine, which computes the appropriate adaptation plans. Though much research has already been done in this domain, the problem of semantic constraints has generally been neglected. This paper addresses the problem of selecting the optimal adaptation operation to satisfy semantic constraints while maximizing the utility of the adapted video. To this end, we define a utility function that computes a value for each possible adaptation operation. We represent our utility function using the MPEG-21 Digital Item Adaptation (DIA) tools. This facilitates the integration of semantic constraints with other types of constraints in MPEG-21 Universal Constraint Description format.

International audience; Web-based social networking services enable people who share interests to find each other and collaborate in online and offline social activities. Thanks to the widespread popularity of networked handheld devices, constantly carried by members of the public throughout their daily activities, new possibilities of collaboration and interaction are emerging, among people who are not only socially close to each other, but also in physical proximity. However, a challenge arises as to how to enable people with similar interests to find each other, so to fulfill their social activities anywhere and anytime, while constantly moving around. In this paper, we present ADESSO, a semi-distributed directory and matching service that supports opportunistic social networking in delay tolerant networks.ADESSO consists of a set of self-organising brokers, automatically elected based on their mobility patterns. Users offload their requests to perform social activities onto brokers upon encounters; brokers then collaborate, by means of either request exchanges or broker fusion, in order to match activities in a way that satisfies users’ social preferences. Preliminary performance evaluation, conducted using real human mobility traces and social networks, shows that ADESSO generates matches that highly satisfy users’ preferences, entailing only a small overhead.

International audience; Information Visualization is a challenging field, enabling a better use of humans' visual and cognitive system, to make sense of very large datasets. This paper aims at improving the current Information Visualizations design workflow, by enabling a better cooperation among programmers, designers and users, in a one-to-one and community oriented fashion. Our contribution is a web-based interface, to create visualization flows that can be edited and shared, between actors within communities. We detail a real case study where programmers, designers and users successfully worked together to quickly design and improve an interactive image visualization interface, based on images similarities.

National audience; Dans cet article, nous nous intéressons à la création et au partage de visualisations d'information. Notre approche est de considérer la visualisation d'information comme le résultat d'un flot de traitement de données, constitué d'un assemblage de services web qui ont vérifié des règles syntaxiques et sémantiques. Afin de faciliter la composition de ces services, et donc de créer des visualisations d'information, nous introduisons mashviz, une interface de programmation visuelle destinée aux concepteurs de visualisations, ainsi qu'aux utilisateurs pour le partage et l'annotation de ces visualisations. Nous discutons les premières visualisations créées et leur partage, puis donnons les prochaines étapes de nos travaux.

In this article, we introduce a new problem: the construction of multi-structured documents. We first offer an overview of existing solutions to the representation of such documents. We then notice that none of them consider the problem of their construction. In this context, we use our experience with philosophers who are building a digital edition of the work of Jean-Toussaint Desanti, in order to present a methodology for the construction of multi-structured documents. This methodology is based on the MSDM model in order to represent such documents. Moreover each step of the methodology has been implemented in the Haskell functional programming language.

International audience; Dans des environnements ubiquitaires, un tiers qui veut faire une transaction exige souvent un certain niveau de confiance envers l'autre tiers. Dans beaucoup de cas, les tiers sont inconnus l'un de l'autre et ne partagent ainsi aucune confiance préexistante. Les systèmes à base de confiance permettent aux utilisateurs d'établir la confiance envers les utilisateurs inconnus grace à des recommandations de confiance des utilisateurs connus. Par exemple, Bob peut vouloir se fier à l'utilisateur inconnu Carol quand il reçoit la recommandation de son amie Alice que la fiabilité de Carol est 0.8 sur l'intervalle [0,1].Dans cet article nous traitons le problème suivant : quand une valeur de confiance est recommandée par un utilisateur à un autre, elle peut perdre son sens réel en raison de la subjectivité. Bob peut considérer 0.8 comme une très haute valeur de confiance mais il est possible qu'Alice ait perçu cette même valeur comme seulement une valeur moyenne. Nous présentons une solution permettant l'élimination de la subjectivité dans la recommandation de confiance. Dans un web de confiance basé sur un graphe aléatoire, nous avons remarqué que notre méthode donne de meilleurs résultats jusqu'à 95% du temps.