A novel approach to sentence clustering

Sentence clustering is often used as the first step in various information retrieval tasks like automatic text summarization, topic detection and tracking etc. Researchers face difficulty to cluster sentences because a single sentence is less informative compared to document. We present a sentence Feature Based Sentence Clustering, FBSC, which incorporates some sentence level relationship features like transition relationship, anaphoric relationship, and term (word) similarity relationship in association with Markov Clustering Algorithm (MCL) to generate sentence clusters. We observe that Purity of clusters generated by Feature Based Sentence Clustering (FBSC) compared to baseline k-means sentence clustering is better.