Abstract:
The emergence of social media has revolutionized the Web, notably by allowing users to interact,
exchange messages and share their knowledge with other users in the form of comments, annotations and
ratings of resources. These tasks have led to a dramatic growth of information on the web. This new
information, known as social information, has been a source of evidence, in the field of social information
research, for estimating the relevance of documents and better responding to user requests. However, the use
of social information to improve information retrieval has several challenges, the most important of which are
(i) find a better representation of documents taking into account the social dimension, (ii) adapt the models of
information retrieval to take into account the different types of social information such as comments and
annotations, (iii) find a better representation of the user request which is generally complex and formulated in
natural language in social forums. The main contributions of our work consist in proposing an approach based
on reduction and expansion to process natural language queries and better understand user needs. We also
propose to adapt and parametrize the IR models to suit the different types of social information. Finally, in
order to better exploit users' reviews, we propose a new representation of documents, which combines terms
and features extracted from user reviews. The proposed approaches were evaluated on two datasets, Social
Book Search and App Retrieval, and the results clearly show the improvement in search performance.
