Query Expansion Based On User Annotating Information

Query expansion technology are developed to solve the problem that the queries users submit can’t describe their truly information need and retrieval intention. Nowadays, all kinds of external extension resources are introduced as the source of the query expansion, such as Wikipedia, query logs and so on. As the social annotation system developed, researchers started to explore the possibility of the social annotation serving as the source of expansion terms. At present, although the good performance of the social annotation based query expan-sion method has been verified, there is still some room to improve the effectiveness. This pa-per aims at achieving better expansion terms in the social annotation system.In the social annotation system, the quality of users’annotating affects the relevance of the annotations and the resources directly. Therefore, the annotations tagged by high quality users are more likely being good expansion terms. Moreover, learning to rank method can achieve better rank through integrating valuable features. Based on the analysis above, the contributions of this paper can be summarized as following two aspects:First, we propose a user quality mining algorithm, which is merging the user tagging be-havior information and the mutual reinforcement principle to assign a reasonable quality score for each user. Based on the user quality mining algorithm, we propose two kinds of query ex-pansion methods with the user quality information added in two modes:(1) obtaining the ex-pansion terms from the user quality filtered resources;(2) weighting the annotation assigned by the high quality users more in the expansion terms selection procedure.Second, we propose a query expansion framework that optimizes the combination of three query expansion methods to achieve better expansion words, which are the annotation term co-occurrence based query expansion approach and two query expansion approaches merging with the user quality information mentioned above. Furthermore, we also introduce learning to rank methods for phrase weighting, and select the features from social annotation resource for training ranking model.Our experiments show that both the query expansion approaches based on user quality and the query expansion approach learning from multiple expansion strategies can achieve better expansion terms to optimize the query expansion performance.