Development of efficient topic modelling techniques using enhanced semantic patterns

Abstract

In every fields of technology, knowledge is growing very fast in online web repository like Wikipedia, Search engines, or offline in the form of books, newspapers and journals etc. Searching in these huge knowledge sources using traditional keyword based search is not an easy task. So semantic based search or concept based search is very important. Concept is also called main topic of the documents. Topic is represented by words. Like topic Sports- is represented by words -football, stadium, playground, cricket, coach etc. Topic modeling is a technique to extract these topics from large document collections without any external knowledge sources or external help. Topic modeling is an unsupervised technique. Topic modeling techniques are classified as probabilistic techniques and nonprobabilistic techniques. Non-probabilistic techniques broadly cover matrix factorization methods like Latent semantic analysis (LSA) and Non-negative matrix factorization (NNMF). In this research we empirically evaluated both types of topic modeling techniques, but main emphasis was on probabilistic approach of topic modeling. Latent Dirichlet allocation (LDA) is a popular topic modeling algorithm considered as synonym for term topic modeling in the research community. In the first part of this dissertation a detailed comprehensive survey on topic modeling techniques has been done. In this study, we synthesize and analyze approximately 150 articles on topic modeling and present a comprehensive review of topic modeling methods that includes classification hierarchy, Topic modeling methods, Posterior Inference techniques, Taxonomy of different evolution models of latent Dirichlet allocation (LDA ) and applications in different areas of technology including Scientific Literature ,Bioinformatics, Software Engineering, Social network ,and humanities. In the end study is concluded with detailed discussion on challenges of topic modeling... newline newline newline