A model-driven approach for context-aware recommendation

Information overload on the Web has created enormous challenges to users searching for relevant information, goods or services. Moreover, online businesses are often overwhelmed by the complex, but rich, data they have in their information systems and find it difficult to target consumers with the appropriate content. In this paper, we propose a statistical approach for online goods and services recommendation. The recommendation model inspires from consumer psychology and relies on several factors influencing individuals' interests and purchase decisions such as consumers' demographics and intentions, items properties and contextual information. Recommendations are generated using a discriminative model which evaluates consumers' purchases probabilities based on a set of observed variables. In this work, two variants of the proposed recommendation model are detailed and evaluated on different datasets of consumers' navigations and purchases.