A Markov Random Field Topic Space Model for Document Retrieval

TL;DR

This paper introduces a Markov Random Field-based model for document retrieval that improves upon traditional methods by capturing term-document relationships probabilistically and reducing dimensionality with SVD.

Contribution

It presents a novel MRF-based framework for document retrieval that extends LSA with probabilistic dependencies and a new parameter learning method.

Findings

Effective retrieval from large datasets

Efficient dimensionality reduction using SVD

Improved modeling of term-document relationships

Abstract

This paper proposes a novel statistical approach to intelligent document retrieval. It seeks to offer a more structured and extensible mathematical approach to the term generalization done in the popular Latent Semantic Analysis (LSA) approach to document indexing. A Markov Random Field (MRF) is presented that captures relationships between terms and documents as probabilistic dependence assumptions between random variables. From there, it uses the MRF-Gibbs equivalence to derive joint probabilities as well as local probabilities for document variables. A parameter learning method is proposed that utilizes rank reduction with singular value decomposition in a matter similar to LSA to reduce dimensionality of document-term relationships to that of a latent topic space. Experimental results confirm the ability of this approach to effectively and efficiently retrieve documents from substantial data sets.