Research On Query-focused Multi-document Summarization

With the rapid growth of textual documents on the internet,people's requirements for efficiency of information retrieval are getting higher and higher.How to efficiently obtain information from the Internet has become an issue that needs attention.Search engines are the most effective tool for retrieving information and collecting Internet data.However,the results returned by search engines still contain a lot of interference and redundant information.Further analysis and generalization are needed to grasp the key points returned by search engines.Query-focused summarization task can solve this problem.Query-focused summarization(QFS)models aim to generate summaries from source documents that can answer the given query.The cooperation of search engines and query-focused summarization can greatly improve the efficiency of information retrieval.The main research topic of this paper is query-focused multi-document summarization.The research is divided into three aspects: model,data,and application.We proposed a query-focused multi-document text summarization model based on Graph Neural Network,explored weakly supervised methods for query-focused text summarization,constructed an open domain information retrieval system based on search engines.We introduce a heterogeneous graph neural network,and use the information transfer in the graph network to effectively promote the exchange of information between sentences.Experiments on the QBSUM dataset show that the query-focused multi-document summarization model based on Graph Neural Network proposed in this paper has achieved excellent results.We explore weakly supervised methods for query-focused summarization.First,we analyzed the data scarcity problem of query-focused summarization.Then,we compare the similarity of the tasks and use the span-extraction task and the summarization task to obtain weak supervision signals of the query-focused summarization.We train query-focused summarization model on the weakly supervised datasets and test these models on the QBSUM dataset.The experimental results show that the different weakly supervised methods used in this article have achieved certain results.Using the query-focused summarization model based on the results returned by the search engine,combined with the document retrieval module and paragraph ranking module,we develop an open domain information retrieval system.The system accepts questions entered by users,uses search engines to search for relevant web pages,and returns streamlined results after processing,which improves the efficiency of information retrieval.On the basis of the open domain information retrieval system,we developed a research report generation system and a rumor analysis system based on different functional modules,and deployed them on the cloud server.