Multi-Document Summarization Generation And Application Based On Domain Knowledge Graph

With the development of Internet technology,more and more people begin to obtain information from the Internet.However,how to obtain information efficiently in the information explosion era has become an important problem for people to solve.The multi-document summarization technique is an effective way to solve this problem.It generates a summary of these documents by inputting multiple documents,which makes it easier to obtain information.However,there are still many difficulties in multi-document summarization techniques,such as the evaluation of information importance,the filtering of redundant information,the aggregation of fragmented information,the organization of multi-document information,and so on.This paper proposes a multi-document summary generation method based on domain knowledge graph to solve these problems.The main research work is as follows:(1)Research on topic sentence generation of single document.In this paper,BiGRU deep learning framework is adopted as the model.A large number of documents and topic sentences are input into the encoder and decoder of deep learning,and attention mechanism is added to prevent information loss.After training and learning,the model is obtained.When decoding,Beam Search algorithm is added to generate the topic sentence of each document.In this paper,the evaluation score of Rouge was 29.2 on the open corpus of LCSTS,and 35.8on the Civil-Military Integration corpus.(2)Research on the generation of multi-document summaries.This method uses domain knowledge graph to establish the mapping relationship between each topic sentence and the knowledge graph by calculating the semantic similarity between the topic sentence and the knowledge graph node,and then relies on the logical structure of the knowledge graph to reasonably organize the topic sentences and generate multi-document summaries.Experiments were conducted in the field of Civil-Military Integration,and the consistency,non-redundancy and readability of the multi-document summarization were evaluated by manual evaluation method,and the results were all above 3.5 points.Finally,the experimental results show that the use of BiGRU,attention mechanism and Beam Search algorithm can effectively improve the generation effect of topic sentences.The introduction of knowledge graph can improve the consistency,non-redundancy and readability of the results generated by multi-document summarization.