Query Refinement using Latent Dirichlet Allocation

Abstract

This study introduces a query refinement technique
that leverages Latent Dirichlect Algorithm ( LDA ) and coher
ence techniques. The proposed method falls under unsupervised
learning, firstly preprocessing large text corpus, applying topic
modelling via LDA and finding the optimal number of topics
using coherence scores eventually leading to the topics which
are meaningful and relevant to the text corpus. We made use
of coherence measures to make sure the topics extracted from
the corpus are relevant to each other and understandable. The
results demonstrated that the method was successful in topic
extraction and modelling of large text corpus. However, the
results significantly vary on the quality of text corpus on which
the method is being applied.