Tamil Document Summarization Using Laten Dirichlet Allocation

This paper proposes a summarization system for summarizing multiple tamil documents. This system utilizes a combination of statistical, semantic and heuristic methods to extract key sentences from multiple documents thereby eliminating redundancies, and maintaining the coherency of the selected sentences to generate the summary. In this paper, Latent Dirichlet Allocation (LDA) is used for topic modeling, which works on the idea of breaking down the collection of documents (i.e) clusters into topics; each cluster represented as a mixture of topics, has a probability distribution representing the importance of the topic for that cluster. The topics in turn are represented as a mixture of words, with a probability distribution representing the importance of the word for that topic. After redundancy elimination and sentence ordering, summary is generated in different perspectives based on the query. KeywordsLatent Dirichlet Allocation, Topic modeling