International audience

Group recommendation has received increased attention over the past decade. The fundamental challenge in group recommendation is how to aggregate the preferences of group members to select a set of items maximizing the overall satisfaction of the group. Different aggregation methods with different semantics have been proposed. In this paper, we explore a novel semantics of group recommendation, that is, probabilistic majority rule, allowing group members to make a "democratic" decision on which items are appropriate. Specifically, we propose a probabilistic model that captures the probability that a given item satisfies the majority of the group. We show that the naive strategy for computing such a probability is exponential time complexity, and propose an efficient dynamic programming approach to avoid this shortcoming. Furthermore, we design and develop an efficient algorithm, which leverages effective pruning techniques, for recommending the k items with the highest majority satisfaction probabilities. Finally, we demonstrate both the retrieval effectiveness and the efficiency of our approach through extensive experimental evaluation on real datasets.

APIs allow companies to export, via the Internet, their skills and know-how, or even to open up new markets and new media for sale. But to fully exploit the advantages of these services, customers, mainly developers, must be equipped with tools giving the possibility of being able to assemble different services together. Fortunately, the notion of service composition is quite advanced, and different tools exist to compose services. However, as APIs with similar functionality are expected to be provided by competing providers, the key challenge is to find the most relevant compositions. This issue has been addressed in the context of QoS-based composite service selection. The downside, in practice, customers choose services based on the number of call limits. In this paper, we propose an approach to select the most relevant compositions based on the notion of call limit. Specifically, we show how the call limits of the individual services can be aggregated to obtain the call limits of a given composition. Then, we introduce the notion of minimal budget skyline, which comprises the most interesting compositions that fit within the customer's budget. In addition, we develop two algorithms, based on effective pruning strategies, to efficiently compute the minimal budget skyline. Finally, we present a thorough experimental evaluation of our approach.

Location-based recommendation tools assist users in discovering attractive Points-of-Interest (POIs). Next POIs recommendation is of great importance and benefit. In this paper, we propose an attention-CNN based network named Deep-POIs for next POIs prediction by modeling several features such as spatio-temporal, sequential, social and context features. In Deep-POIs model, users’ mobility behavior is modeled as images and then a CNN architecture is employed to capture the spatio-temporal patterns considering the social relations connecting users. An attention mechanism is then employed to extract the sequential features which are essential to describe the user behaviors. The model also integrates context feature extraction. We evaluate our model on two real location-based networks, Foursquare and Gowalla, and show that it significantly outperforms eight existing methods.

Service selection is a challenging task, and a lot of effort has been devoted to tools that assist the user in choosing the service whose non-functional parameters better match her/his preferences. In many practical situations, the responsibility to decide which is the appropriate service is shared among multiple parties. A standard approach to this service selection problem is to discard services that are unanimously considered inappropriate and focus on the rest. However, as the involved parties may have colliding interests, only a few services may be eliminated. This work addresses this shortcoming and enables users to reach a “democratic” decision by means of a majority vote: a service is eliminated if the majority of the parties find it inappropriate. We formulate the problem using dominance relationships, and propose algorithms that return an appropriate subset of services for the parties, while being more efficient than standard techniques. Moreover, we consider the problem of defining an appropriate ranking for the non eliminated services, and formulate it as an instance of a group recommendation problem. Finally, we demonstrate the effectiveness and the efficiency of our approach through extensive experimental evaluation on real-based and synthetic datasets.

Real-time bidding (RTB) has recently become the predominant technique in online advertising. Although, RTB is very effective, compared to classical approaches, a lot can be done to improve the accuracy of display advertising. In fact, a major drawback of existing RTB systems is the use of the bidding price and the current user profile as the primary features to display advertising. However, in doing so, the user may not always get the appropriate ad, and the same ad may be presented several times to the same user, which leads to its frustration. To overcome this limitation, we propose in this paper, an approach to serve the right ad to the right user at the right time. Our approach consists of incorporating the notion of recommender systems into the RTB architecture. Specifically, we design a two-layer approach. The first layer implements the item-based collaboratif filtering technique, while the second layer implements the factorization machines model. This allows to capture and use the information of other users (more specifically, those who are similar to the current user) to enhance the accuracy of display advertising. We show how these two layers collaborate to reach our goal, and validate our approach through an experimental study.

Nowadays, due to the growth of technology, there is a mass production of data (of large volume), available in a digital form. These currently available data are not unified but appear in different formats and types. The diversity of the data is based on the type of information, they contain, such as text, image, video and audio documents and also on their sources, such as data from sensors (high variety). In addition, with the expansion of the Internet and the World Wide Web, the majority of these data become the publicity available for a wide range of users (at high speed). The main objective of this work is to propose an efficient Big Uncertainty Web Data Services Indexing Model able to reasoning in uncertain data environment. More concretely, the proposed approach is based on two main phases: the first one consists on processing uncertain data in the syntactic indexing phase and the second one consists on the semantic indexing phase. These two phases are presented as two algorithms syntactic and semantic.

Data-driven Web services build on service-oriented technologies to provide an interoperable method of interacting with data sources on top of the Web. Data Web services composition has emerged as a flexible solution to answer users’ complex queries on the fly. However, as the number of Web services on the Web grows quickly, a large number of candidate compositions that would use different (most likely competing) services may be used to answer the same query. User preferences are a key factor that can be used to rank candidate services/compositions and retain only the best ones. In this article, we present a novel approach for computing the top- k data service compositions based on user preferences. In our approach, we model user preferences using fuzzy sets and incorporate them into the composition query. We use an efficient RDF query rewriting algorithm to determine the relevant services that may be used to answer the composition query. We match the (fuzzy) constraints of the relevant services to those of the query and determine their matching degrees using a set of matching methods. We then rank-order the candidate services based on a fuzzification of Pareto dominance and compute the top- k data service compositions. In addition, we introduce a new method for increasing the diversity of returned top- k compositions while maintaining as much as possible the compositions with the highest scores. Finally, we describe the architecture of our system and present a thorough experimental study of our proposed techniques and algorithms. The experimental study demonstrates the efficiency and the effectiveness of our techniques in different settings.

Crowdsourcing platforms provide powerful means to execute queries that require some human knowledge, intelligence and experience instead of just automated machine computation, such as image recognition, data filtering and labeling. With the development of mobile devices and the rapid prevalence of s

In this paper, we introduce a novel concept called collective skyline to deal with the problem of multiple users preferences. We then conduct a set of experiments that demonstrate the effectiveness of the introduced concept.

Nowadays, the exploding number of functionally similar Web services has led to a new challenge of selecting the most relevant services using quality of service (QoS) aspects. Traditionally, the relevance of a service is determined by computing an overall score that aggregates individual QoS values. Users are required to assign weights to QoS attributes. This is a rather demanding task and an imprecise specification of the weights could result in missing some user desired services. Recent approaches focus on computing service skyline over a set of QoS aspects. This can completely free users from assigning weights to QoS attributes. However, two main drawbacks characterize such approaches. First, the service skyline often privileges services with a bad compromise between different QoS attributes. Second, as the size of the service skyline may be quite large, users will be overwhelmed during the service selection process. In this paper, we introduce a new concept, called alpha-dominant service skyline, to address the above issues and we develop a suitable algorithm for computing it efficiently. Experimental evaluation conducted on synthetically generated datasets, demonstrates both the effectiveness of the introduced concept and the efficiency of the proposed algorithm.