Exploiting Discourse Relations between Sentences for Text Clustering

Over the years, the usage of discourse relations has been proven to enhance many applications such as text summarization, question answering and natural language generation. This paper proposes an approach that expands the benefit of discourse relations for natural language processing from a different aspect. We exploit the discourse relations existing between sentences to generate clusters of similar sentences from document sets. We first examined and defined the type of discourse relations that useful to retrieve sentences with identical content. We then assigned these relations to each sentence pair using a machine learning method. Finally we performed discourse relation-based clustering algorithm to generate clusters of similar sentences. We evaluated our method by measuring the cohesion and separation of the clusters and compared to a well recognized clustering method. The experimental result shows that our method performed significantly well, which demonstrated that discourse relation between sentences can be exploited for text clustering.