Research on Query Expansion Based on Deep Learning

Abstract

In order to overcome the problem of low retrieval efficiency. The paper proposes a method of deep learning-based distributed vector representation for query expansion. The classic task of improving queries to improve retrieval performance is query expansion, refining user intent by filling in extended terms to fully understand the needs to achieve retrieval accuracy. In the query expansion, how to select the extended words is the key issue, and the quality of the extended words determines the performance of retrieval. The main feature of this method is to optimize the expansion words, improve the extended word relevance labeling strategy based on learning to rank, and use the word vector to construct features for the extended words for the construction and optimization of the extended word ranking model. Experimental results show that the method has a high accuracy rate on TREC public dataset, with a 4.45% improvement compared to the traditional method, which has important implications for the research of deep learning in information retrieval.