Personalized query expansion utilizing multi-relational social data

Social tagging systems have been widely used as a way to annotate and categorize Web resources. However, users often use unrestricted vocabulary to tag and describe resources. On the contrast, annotators of Web documents may use very different words to describe the same concept. In the past few years, numerous personalized query expansion methods have been proposed to tackle the vocabulary mismatch problem. Many of them are based on the probabilistic-based techniques or graph-based techniques, but they ignored the multi-relational characteristics existed in the social data. In this paper, we explore multiple semantic relationships from social tagging systems, including relationships between tags, between words and between tags and words. Three affinity graphs are built based on the features derived from tags and words. In addition, we incorporate pseudo-relevance feedback information obtained from top-ranked documents to regularize the smoothness of multiple associations over the three affinity graphs. The key of this paper is considering above three affinity graphs into a novel query expansion model and aim to produce better personalized search results. Experiments conducted on a real-world dataset validate the effectiveness of the proposed approach.