Book search using social information, user profiles and query expansion with Pseudo Relevance Feedback

Abstract

Book Search has gained astounding popularity worldwide. Nowadays, users search the items/products online. Users who have not any idea about the product they look towards the social information and user profiles. Social information is further categorized into structured information (e.g. rating and tags) and unstructured information (reviews and annotations). Consequently, how to offer the best recommendation or suggestion of items to end users is becoming a hot topic among researchers. The retrieval and recommendation of relevant documents to the users is a key issue in many domain e.g. songs, accessories, movies, books, etc. In this paper, taking social books as an example, we propose a novel Pseudo Relevance Feedback (PRF) framework for retrieving and searching for relevant documents using social information and user profiles. Especially, we have redesigned a typical distribution-based term selection strategy and transformation-based term selection strategy. Terms are selected and weighted in hope to avoid word mismatch problem and to improve retrieval of the relevant document. Finally, we develop a searching system, where Learning-to-Rank technique is used to adaptively combine the results which are obtained from various PRF strategies with user profiles and social information. Our proposed methodology is extensively evaluated on INEX/CLEF Social Book Search Track (SBS) datasets to verify the effectiveness and robustness of the proposed method. As a result, our proposed method shows the best performance (nDCG@10) on all 3-years SBS track (Suggestion Task) datasets compared to other state-of-the-art methods.