Comparison of Topic Language Models for Query Disambiguation in Information Retrieval

A long-standing challenge in information retrieval is to disambiguate query words for more precise search results. However, two or more meanings of a word in a query, or polysemy, deteriorate the precision effectiveness of information retrieval systems. There is a need for correct and effective information retrieval in many information systems such as health care and customer relationship management. This paper examines three topic language models that are mentioned in the literature for their ability to handle polysemy in query words. The three topic lanauge models are--latent semantic analysis, probabilistic latent semantic analysis, and latent Dirichlet allocation. We review these models and compare their performance in query disambiguation. Our study provides guidance on the use of these models in information retrieval systems.