Diversifying Search Result Leveraging Aspect-based Query Expansion

Abstract

Web search queries are short, ambiguous and tend to have multiple underlying interpretations. To reformulate such queries, query expansion is a prominent method that leads to retrieve a set of relevant documents. In this paper, we propose an aspectbased query expansion technique for diversified document retrieval. At first, query suggestions and completions are retrieved from major commercial search engines. A frequent phrase-based soft clustering algorithm is then applied to group similar retrieved candidates into clusters. Each cluster represents different query aspect. The expansion terms are selected from the generated cluster labels for each cluster. To estimate the relevancy between the expanded query and the documents, multiple new lexical and semantic features are introduced using the content information, and word-embedding model, respectively. Finally, a linear ranking approach is employed to re-rank the documents retrieved for the original query using the extracted features. We conduct experiments on Clueweb09 document collection using TREC 2012 Web Track queries. The experimental results clearly demonstrate that our proposed aspect-based query expansion method is effective to diversify the retrieved documents and outperformed baseline and some known related methods in terms of diversity metrics ERR-IA, α-nDCG and NRBP at the cut of 20.