Abstract—The coming of the Digital TV will bring a significant increase in the number TV programs offered by TV operators. Consequently, the user are facing it difficulty to find out the most interesting TV programs among the various options available. In this new scenario, the recommender systems stand out as a possible solution to the information overload problem. However, the current approaches to recommend content for Digital TV rarely considers the context during the recommendation process. Thus, this paper presents a software infrastructure – entitled PersonalTVware - to support context-aware recommendation of TV programs. To validate the PersonalTVware, a context-aware recommender system was implemented as a concept proof. In order to evaluate the quality of the recommendation, a number of experiments have been conducted. The results indicate that consider both user’s profile and context can provide better recommendations. Index Terms—Context-awareness, interactive digital TV, recommender systems, ubiquitous computing. I. INTRODUCTION The advent of Digital TV will bring a significant increase in the number of TV programs available to end-users, which leads to information overload problem [1]. Consequently, the users are facing this problem and having difficulties to find out the favorite TV program among the options available. The traditional tool known as Electronic Program Guide (EPG), therefore, has not efficiently responded to the user’s needs. The EPG simply displays long lists of TV programs requiring the user to spend a great deal of time looking for information on his/her favorite TV programs. In this new scenario, the recommender systems stand out as a possible solution. These systems filter relevant items according to user’s preferences or group of users who have similar profiles. However, the current generation of recommender systems for Digital TV operates in two-dimensional User x Item space. That is, they make their recommendations based only on the user’s profile and item information without taking into consideration the contextual information [2]. In many situations, the user's interest may also depend significantly on the context. According to Dey [3], the context is any information that can be used to characterize the situation of Manuscript received February 16, 2012; revised March 27, 2012. Authors are with the Laboratory of Computer Architecture and Network, University of São Paulo (USP), São Paulo, Brazil (e-mail: fsilva@larc.usp.br; luizgpa@larc.usp.br; gbressan@larc.usp.br). 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 the application themselves. In the case of recommender systems for Digital TV, it is also important to consider the information about the context of the entity user. Generally, the contextual information can be identified from six basic contextual known as: where (location), who (identification), what (action or activity), when (time), why (motivation behind the action) and how (a way to identify how the elements of the context are collected) [2]. According to Dey [3] a context-aware system uses the context to provide relevant information and/or services to the user, where relevancy depends on the user’s task. Thus, it becomes important to extend the traditional approaches for automatic content recommendation to support the exploration of the user’s context, which may improve the quality of the generated recommendations [2]. Some questions related to the context can be exploited such as: who is the user and when he attends a particular genre of TV programs? On Sunday morning or on Monday evening and when does he come from office? Where and how the TV program will be seen? At home through a Digital TV receiver connected to full HD TV set or at school on portable TV? And what kind of TV program is considered relevant in such situation for the user then watching TV? Depending on his/her context, the user may have different viewing preferences and needs. Therefore, this paper presents the PersonalTVware, a software infrastructure to support the development of context-aware recommender systems for Digital TV. The PersonalTVware provides components that allow the TV programs filtering, manage information regarding the context, user profile and TV programs, and cross-context reasoning to infer contextual preferences. The task of inferring contextual preferences is based on data mining and machine learning techniques such as decision tree classifier, naïve Bayesian classifier, back-propagation (a neural network), and case-based reasoning technique [4]. The context-aware information filtering is based on content-based filtering technique [2]. Thus, developers of recommender systems focus efforts on usability concepts of their systems, leaving questions on the low-level PersonalTVware manage. To validate the PersonalTVware in a case study, a context-aware recommender system was implemented as a concept proof. This paper is organized as follow: section II discusses the related works, section III describes the PersonalTVware PersonalTVware: An Infrastructure to Support the Context-Aware Recommendation for Personalized Digital TV Fábio Santos da Silva, Luiz Gustavo Pacola Alves, and Graça Bressan International Journal of Computer Theory and Engineering Vol. 4, No. 2, April 2012 131