Aspect Based Multi-Document Summarization

Multi-document summarization is useful when a user deals with a group of heterogeneous documents and wants to compile the important information present in the collection, or there is a group of homogeneous documents, taken out from a large corpus as a result of a query. We present an approach to automatic multi-document summarization that depends on clustering and sentence extraction. User provides a query, based on the query; documents that are relevant to the query are extracted from a document corpus containing documents from various domains. An n x n similarity matrix is created among the sentences having sentence level similarity in all extracted documents. Then clusters of similar sentences are formed using Markov clustering algorithm. In each cluster, each sentence is assigned five different weights 1. Chronological weight of sentence (Document level) 2. Position weight of sentence (position of sentence in the document) 3. Sentence weight (based on term weight) 4. Aspect based weight (sentence containing aspect words) and 5. Synonymy and Hyponym Weight. Then top ranked sentences having highest weight are extracted from each cluster and presented to user.