A Contextual Query Expansion Model using BERT Based Deep Neural Embeddings

Abstract

The amount of information available on the internet is growing exponentially. The majority of this information is ambiguous by nature, and information retrieval (IR) systems typically return unrelated information when a typical web user tries to find relevant data. In this paper, we proposed a contextual query expansion technique (CQEB), which allows us to select only relevant documents and then only relevant terms from those documents. In order to establish the connection between retrieved documents and query keywords, the CQEB method makes use of BERT based deep neural word embeddings. We compared CQEB with the Glove embedding based QE technique. Extensive testing on test datasets from CACM and CISI reveals that our suggested method, CQEB, performs better than the standard query expansion (QE) techniques. Our experimental analysis demonstrates that, in 96% of the cases, the proposed method CQEB outperforms the alternative strategies in terms of F-score.