Enhancing Latent Semantic Analysis by Embedding Tagging Algorithm in Retrieving Malay Text Documents

Abstract

Latent Semantic Analysis (LSA) is a mathematical approach that uses Singular Value Decomposition to discover the important association of the relationship between terms and terms, terms and documents and also documents and documents. LSA adopted cosine similarity measure to calculate the similarity between the query and terms as well as the documents. This approach seem to be efficient if each of the term only have single meaning and a meaning only represent by single term. Unfortunately, there are terms that have multiple meanings and a single meaning that represent by multiple terms. If these terms are treated as a single word, it will lead the search engine to retrieve the irrelevant documents. The irrelevant documents that been retrieved will affect the effectiveness of the search engine. This paper propose to enhance LSA by embedding tagging algorithm. To investigate the effectiveness of LSA using tagging algorithm in retrieving documents from the Malay corpus, seven experiments are conducted. The first experiment conducted to compare the time taken for extracting the normal term list and the tagged term list, total number of both lists and also the time taken for the creation of term document matrix. All other experiments record all the results of the retrieval system by using different dimension and threshold value. The retrieval result averagely shows F-measure enhancement of approximate to 3.60% by using LSA with tagging algorithm (LSAT) compared to retrieval result of LSA.