The Impact of the Intrinsic context on the Recommendation Process Latifa Baba-Hamed #1 , Réda Soltani #2 # RIIR Laboratory, University of Oran Oran-Algeria 1 lbabahamed@yahoo.fr 2 reda.soltani@hotmail.fr Abstract — In this paper we present a new approach which allows the introduction of intrinsic context in the recommendation process. In this approach, we define the context from the point of view of objects. We show, also, by combining it with the user's profile, how it can improve the accuracy of the recommendation and thus better meets the needs of users. Keywords— Recommendation, context, user profile, preference, matching operator, precision. I. INTRODUCTION The popularization of the Internet and the explosion of recommendation services today have propelled the information retrieval (IR) in the foreground. Indeed, the overabundance of information has led to the deterioration of the quality of results returned by the web to users. Recommender systems (RS) are effective tools to overcome the problem of information overload by providing users with relevant contents. Relevance is measured by the similarity between content and user profile which consists of a set of preferences. A more complete definition of the user profile is presented in [17]. RSs operate according to three filtering strategies, namely, content-based filtering CBF (which consists of matching between a user profile and content descriptors, in order to recommend appropriate products) [18, 26], collaborative filtering CF (which assumes that users who had common interests in the past, will continue, probably, to share the same tastes in the future) [10, 13, 20], or hybrid filtering (which is a combination of these two approaches) [23]. RSs traditionally operate on a user-item matrix. As the user enters new ratings or makes new purchases, their user profile is updated by simply adding the new information to the current rating vector for the user. These systems ignore the fact that users interact with systems within a particular “context” and ratings for items within one context may be completely different from the rating for the item within another context. One of the most frequently cited definitions of a context was proposed by Abowd, Dey et al [2]: « Context is any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves ». Several context-aware systems have emerged recently. Among these works, we can mention: Micro-profiling [8] (an approach which assumes that users’ preference changes over time but has a temporal repetition. The main idea of this approach is to replace the single user profile by taking into account many specialized profiles each representing the users' in different contextual conditions), DaVI [12] (an approach that represents the context as a virtual item. for example the attribute day Di = {1, .., 31} can be considered as a virtual and contextual object, in computing recommendation), News @ hand [9] (is a news recommendation system which considers the semantic contextualization), MyMap [11] (use a map on mobile phone to provide personalized recommendations of places of interests for tourists in a city), DISCOVR [14] (system which consists of three different components in the form of separate services: Sensor, recommendation and utility services. A sensor is a service that acquires information about the user's context), Sourcetone [25] (the system asks the user's emotional state before recommending songs), Amazon [4] (a system which is specialized in selling books and other products, it provides a button "find gifts" that allows it to distinguish between user-specific preferences and the preferences of a person to whom the user wants to give a gift), PAM [1] (is a personalization system in the context of content delivery platforms, it is inspired by [3] and allows the simultaneous consideration of user profiles and their contexts). A detailed state of the art and a comparison of the main context-aware systems are presented in [24]. All the works cited above were restricted to the extrinsic context (context for the environment of user interaction). Very few researchers have examined the intrinsic context (context being part of the object itself). In this paper, we present a new approach for the consideration of intrinsic contextual information and its impact on the recommendation process. We can apply this approach for the recommendation of contents in overall (book, Url, article, product, movie, song, restaurant, etc.). For the illustration and the evaluation, we chose the field of movies. The rest of the paper is organized as follows. In Section 2, we recall the data model and the formalism followed. Section 3 is devoted to the architecture of the contextual system and its main components. An illustrative example in the case of films recommendation is given in Section 4. In Section 5, we present the results of the evaluation of the developed system.