Research On News Event Retrieval Technology Based On Feature Graph

Nowadays,the Internet is inseparable from our lives.Netizens are very accustomed to searching through the Internet to understand the big news and events happening in the world and find the information they care about.However,the rapid development of technology has led to the rapid spread and explosive growth of information,and various types of information have become increasingly fragmented.When people want to inquire about a news event of interest to understand the complete context and evolution of the event,due to the timeliness of the news,each media will only report when the event has an important turning or progress,and most of them only introduce the current event situation,the users need to perform multiple queries by themselves.It is very inconvenient to read many related articles to understand the cause and effect and complete information of the event.Not only that,when the user wants to further expand reading and learn more similar events,the existing search engine does not recommend similar events for the event content comparison,so the user can not obtain the integrated similar event information through the existing retrieval operation.This paper proposes two problems for event information integration and similar events recommendation in the field of news event retrieval.After combing relevant literature and research results,a news event retrieval model based on feature graph is proposed.The model firstly combines the TextRank algorithm with the initial position information of the keyword to form the TR-F algorithm to extract and process the keywords in the news document,and use the conditional random field to identify and extract the place names in the document,and use the news release time to obtain the time information of the event.Then,by calculating the similarity of the event features,the event content clustering forms the event feature graph,and the time series-based event puzzle function is realized.This function can effectively integrate a large amount of fragmented information and splicing into a complete event development context.Then,this paper proposes a method of segmenting event according to time span.By combining the VSM model with the segmentation metric proposed in this paper,the similaryity of event content is calculated and the BM-25 algorithm is improved to achieve efficient retrieval of query questions and efficiently recommend other events which are similar to the target query event for the user.After that,this paper proves the validity and feasibility of the news event retrieval model based on the feature graph proposed by comparing and analyzing the experiments on the real data set.Then based on the retrieval model,this paper designs a news event retrieval prototype system based on the feature graph,and introduces the function and implementation strategy of each module in the system.