Sparse Representation Based Query Classification Using LDA Topic Modeling

In recent years, tremendous growth of documents provides scope and challenges to the interdisciplinary research community in text processing for retrieving information. Text analytics reveals high-quality information by identifying patterns and its trends using statistical methods. In this paper, we propose a novel approach to classify user query in a reduced search space by considering the query as a collection of words distributed over different topics. Latent Dirichlet allocation (LDA) has been used for topic modeling and a collection of topics containing words are obtained following Dirichlet distribution. We construct a sparse matrix called topic-vocabulary matrix (TVM) using probability distribution of words appearing in the topics. Finally, sparse representation based classifier (SRC) has been applied for classifying query using TVM consisting of training patterns. Here, we have analyzed the effect of number of patterns in classifying the queries and achieved 90.4 % accuracy.