Personalized Semantic Query Expansion Based on Dynamic User Query Profile and Spreading Activation Model

Shengrong Zhu, Xiangguang Meng, Feixiang Chen, Xuan Tian
2017 International Journal of Hybrid Information Technology  
Semantic query expansion is a widely used method to resolve the query problems of synonym and polysemy in the information retrieval field. However, it does not make users more satisfied with the search results because too much noise unfit to users' needs is introduced in the process. In this paper a new framework combining personalization with semantic query expansion is proposed to overcome the noise problem brought by semantic query expansion. In the proposed framework, firstly, instead of
more » ... stly, instead of using traditional hierarchical expansion strategy, the spreading activation model (SAM) is used for enhancing the selection of expansion terms to reduce the noise. Secondly, to get more accurate expansion terms for individual search, dynamic user query profile is built to capture individual variable query needs and is integrated into the semantic expansion process. The proposed expansion process is described by four steps: building dynamic user query profile, concepts mapping, personalized semantic query expansion and determining the final expansion terms. Four groups of experiments were designed to verify the validity of the proposed method. The experiment results show that the proposed method outperforms both traditional hierarchical expansion and keyword-based query, which manifests that building dynamic user query profile is important for depicting user query needs in semantic query expansion and it is more rational to improve query expansion based on spreading activation model. Moreover, personalized semantic query expansion based on dynamic user query profile and spreading activation model can reduce noise of semantic query expansion and improve the search effectiveness. Keyword: semantic query expansion, personalized information retrieval, dynamic user query profile, spreading activation model However, semantic query expansion approaches may result in topic drift problem, namely too many noise terms unfit to users' needs are introduced in the process. For example, "program" have multiple paraphrases: television show, plan or computer program. Assuming user tends to search something about computer program, the retrieval without considering semantic problem may get the results only if it contains "program" and including all sense of "program", while an automatic expansion of queries containing "program" with "computer" might work well in this case. That means individual needs can help to exclude noise terms in semantic query expansion. In fact, there have been many researches on semantic query expansion combined with personalization [7, 8, [16] [17] [18] [19] . These researches often study user profiles and personalized search algorithm to enhance the semantic query expansion. However, how to enhance the selection of expansion terms on large scale thesauri or semantic supporter to reduce the noise is often ignored. And another fact is that a search user's needs are also dynamically changed with time, and which is often neglected in the aforementioned personalized semantic query expansion related researches. In this paper, a new framework is proposed to combine personalization with semantic query expansion. A spreading activation model which simulates the thinking process of human brain more vividly is utilized as the expansion strategy to enhance the selection of expansion terms. And a kind of dynamic user query profile is built to capture individual variable query interest to reduce the noise terms unfit to users' needs. The implementation of the framework includes four steps. Firstly, building dynamic user query profile on the basis of query logs; secondly, mapping query terms to appropriate concepts on WordNet; thirdly, exploring spreading activation model to implement personalized semantic query expansion; finally, determining the selection of final expansion terms. In addition, variant systems are built for comparison, like keyword-based searching with default Lucene performance (NE), general hierarchical query expansion without mapping filtration (EHNF), and hierarchical query expansion with mapping filtration (EHF). The evaluation is performed using MAP and the degree of relevance (DOR). The result shows that our proposed method performs well and is better than compared ones. In the rest of this paper, Section 2 reviews related work on two aspects: semantic query expansion and personalized search. Section 3 describes the proposed framework and detailed implementation. Section 4 presents experiment data, experiment design and evaluation result. Section 5 concludes the paper. Experiment results Parameters: During the experiments, we adjust all parameters mentioned in previous sections to optimize the proposed method, the value are given in Table2. Besides, given the query sparsity in query history, the time unit in damping function is set one month.
doi:10.14257/ijhit.2017.10.6.04 fatcat:26tg6yrpnjh5rdacl6o4v24c2m