The Impact of Linked Documents and Graph Analysis on Information Retrieval Methods for Book Recommendation

A new combination of multiple Information Retrieval approaches are proposed for book recommendation based on complex users' queries. We used different theoretical retrieval models: probabilistic as InL2 (Divergence From Randomness model) and language models and tested their interpolated combination. We considered the application of a graph based algorithm in a new retrieval approach to related document network comprised of social links. We called Directed Graph of Documents (DGD) a network constructed with documents and social information provided from each one of them. Specifically, this work tackles the problem of book recommendation in the context of CLEF Labs precisely Social Book Search track. We established a specific strategy for queries searching after separating query set into two genres Analogue and Non-Analogue after analyzing users' needs. Series of reranking experiments demonstrate that combining retrieval models and exploiting linked documents for retrieving yield significant improvements in terms of standard ranked retrieval metrics. These results extend the applicability of link analysis algorithms to different environments.