Semantic Similarity of Documents Using Latent Semantic Analysis

Chelsea Boling, Kumer Das


Latent Semantic Analysis (LSA) is a technique that analyzes relationships between documents and its terms, and it discovers a data representation that has a lower dimension than the original semantic space. Essentially, the reduced dimensionality preserves the most crucial aspects of the data since LSA analyzes documents to find latent meaning in the corpus. The latent semantic space is determined by singular value decomposition (SVD), which enables a powerful process to simplify any rectangular matrix into a product of three unique components. The purpose of using SVD is to retrieve a sufficient amount of dimensions, which reveal a relevant structure that spans the original term-document matrix. In this study, we use LSA to find particular associations with user queries in a sample of MEDLINE documents. Based on our experiments, selecting an appropriate dimension for a reduced representation is suitable to represent the original latent space. The reduced model of the term-document matrix shows that SVD is capable of dealing with semantic problems at a promising cost. Overall, the goal is to overcome the problem of unsatisfactory indexed results by revealing meaningful relationships between terms and documents.


Latent Semantic Analysis, Singular Value Decomposition, Text Mining

Full Text: PDF


  • There are currently no refbacks.