Personalized Information Retrieval Based on Probabilistic Graphical Model for Sparse Data

Faced with the various and massive information resources, it is prominent to provide users with accurate, personalized service content efficiently and comprehensively, addressing their diverse retrieval needs. To this end, for University Digital Libraries (UDLs), we propose a Personalized Information Retrieval method for UDLs based on Probabilistic Graphical Model (PIRPGM). This method integrates both keyword retrieval and semantic retrieval to enhance the relevance and precision of search results. We introduce a “Spike and Slab” prior and design a novel Collapsed Variational Bayesian (CVB) inference algorithm to estimate model parameter. The PIRPGM not only offers deep insights into topic but also alleviates data sparsity. Its effectiveness is verified across different lengths (e.g., short texts and long texts) and scenarios, particularly benefiting cold start users.