Event-based Summarization for Scientific Literature in Chinese

Abstract Large-scaled scientific literature such as papers, patents and reports have been published with high speed by researchers all over the word. It becomes a difficult task for researchers to read all the latest and relevant literature. So there is an urgent need for summarization systems to provide brief and important dynamic information in research domains that researchers are interested in. This paper proposes an approach to generate summarization with 5W1H event structure. Sentences in the literature are classified and selected for different elements of events by relevance, and then the importance of each candidate sentence is calculated. Top-k relevant and important sentences are selected to formulate event-based summarization. Comparing with existing summarization results or written by authors manually, experiment results of our approach have more detailed information with 5W1H event structure, which is more convenient for researchers to search and browse brief description of scientific and technical information.