利用嵌入方法实现个性化查询重构

As a mechanism to guide users towards a better representation of their information need, a query reformulation method generates new queries based on the queries issued by users. To preserve the original search intent, most of the current query reformulation methods are to obtain the context information of query term based on the co-occurrence information between the query terms, and then use the contextual similarities to generate candidate query reformulations. Finally, these candidate query reformulations are scored according to the semantic consistency of terms, dependency among latent semantic topics and users’ preferences. However, we exploit embeddings method to realize the personalized query reformulation. Firstly, we use the query term embedding technique to obtain the vector of each, and this vector represents the contextual information for each term. Secondly, the vector which characterize the users’ preference are constructed by using the tem vectors, and candidate query reformulations are generated according to user preference based on term vectors and user vectors. Finally, topic embeddings are proposed to extract the context information of each latent topic of term, and hidden Markov model (HMM) is used to integrate tem vectors, user vectors and topic vectors to re-rank the candidate query reformulations based on users ‘personalization. The final experimental results show that the method outperforms the existing method.