Query Expansion for Information Retrieval using Word Embeddings: A Comparative Study

Internet in today's times is the daily need of people. To retrieve right information efficiently is the constant desire. Expanding user queries by transforming some keywords to retrieve specific domain keywords has been a probable solution for information retrieval. Various methods have been combined with query expansion from time to time to improve the information retrieval results right from Classical IR methods to semantic methods or to natural language processing methods. All the methods have eventually minimized the mismatch problems and gave better retrieval results. This paper discusses the performance of various such methods that can be implemented to expand user query such that it gives high precision search results. The paper mainly focuses on word embeddings methods like Word2Vec - CBOW or Skip gram and Glove that are trained on real estate related legal datasets over classical methods. Experimental results show that word embeddings give better results with 87% mean average precision (mAP) values on all datasets.