A submodularity based game theory framework for multi-document summarization

The explosion of a large amount of data over the internet has made it inconceivable to effectively forage relevant information from it. To overcome the above problem, a fundamental tool known as summarization techniques have been used for understanding and organizing such large datasets. One approach of summarization widely discussed in the literature is called multi-document summarization, where extracts of texts are selected to best represent the essence of a document. A game theory framework for multi-document summarization is proposed and a combination of three similarity measures is utilized to capture similarity between textual units. The proposed framework ensures that the summary obtained is well covered and informative. Results of the proposed approach were evaluated with ROUGE evaluation metric on DUC and TAC datasets against human-generated summaries and it outperformed other competitive methods.