A Text Summarization Method Based on Semantic Similarity Among Sentences

Abstract

In recent years, more and more attention has been paid to the graph-based text summarization. In terms of this method, the document is transformed into a text graph and then sentences are ranked by taking into account the global information of the document. However, the similarity measure be-tween sentences is mostly limited to word layer in previous studies, which makes it difficult to measure the semantic similarity accurately; besides, the features of sentence has also been disregarded. In view of the defects mentioned above, this paper proposes a semantic-word two layer similarly measure model and defines topic relativity as well as position sensitivity, so as to optimize the sentence ranking results. Through the evaluation on DUC datasets, performance of this approach shows greater improvement compared with a number of baseline systems.