Visual Text Summarization in Supervised and Unsupervised Constraints Using CITCC

Abstract: In this work clustering performance has been increased by proposes an algorithm called constrained informationtheoretic co-clustering (CITCC). In this work mainly focus on co-clustering and constrained clustering. Co-clustering method is differing from clustering methods it examine both document and word at a same time. A novel constrained coclustering approach proposed that automatically incorporates various word and document constraints into informationtheoretic co-clustering. The constraints are modeled with two-sided hidden Markov random field (HMRF) regularizations. An alternating Expectation Maximization (EM) algorithm has developed to optimize the model. NE extractor and WordNet methods are proposed to automatically construct and incorporate document and word constraints to support unsupervised constrained clustering. NE extractor is used to construct document automatically based on the overlapping named entities. WordNet is used to construct word constraints automatically based on their semantic distance inferred from WordNet. It can simultaneously cluster two sets of discrete random variables such as words and documents under the constraints extracted from both sides. With this work contains add visual text summarization to increase more clustering performance.