Font Size: a A A

Hot Topics Detection Analysis And Trend Research On Network News

Posted on:2019-10-18Degree:MasterType:Thesis
Country:ChinaCandidate:Z X ShiFull Text:PDF
GTID:2429330545470814Subject:Applied statistics
Abstract/Summary:PDF Full Text Request
With the advent of the "Internet Plus" era,the Internet has gradually become an important channel for people to obtain information and disseminate information.While a large number of online news sources enrich people's lives,they also contain a great deal of value,such as online news in the public opinion and stock market forecasts.There are important applications.However,network news is staggered and disorganized,and people often cannot obtain timely and effective information.Network news topic detection and trend research can well solve this problem.Network news topic detection is mainly used to detect from massive amounts of online news.Hot topics to facilitate people's attention to social issues.This article makes some improvements based on traditional topic detection methods.First of all,the data in this paper comes from the crawling of the eight categories of web news data of each major portal in January 2018.Then this paper adopts the method of joint modeling of Word2 vec and LDA in the selection of topic models;Before the clustering,text classification is used to preprocess,so that different types of network news can be obtained;then a doublelayer Single-Pass clustering is designed for topic discovery.Finally,the topic of the topic is studied and the topic is discussed.The formula of heat and topic index calculates the trend of topics.According to the results of the research,the combined model of Word2 vec and LDA used in this paper shows excellent results.Experiments show that the performance of the model compared with the single model is significantly improved.However,the classification model based on Word2 vec and convolutional neural networks constructed in this paper has achieved good results.The effect of classification accuracy is over 90%.Secondly,the double-layer Single-Pass clustering designed in this paper has excellent clustering effect and can handle continuous-time network news with strong flexibility.Finally,the hot topics proposed in this paper are presented.The trend research method is compared with the authoritative search engine Baidu Index in the actual topic case analysis,and the findings are almost the same,which confirms the validity of the trend research in this topic.Therefore,through the model of this article,we can detect hot topics from the mass of online news,and study topic trends.This has strong application value to users,enterprises,and governments.
Keywords/Search Tags:Topic detection, Trend research, Word2vec, topic model, Double Layer Single-Pass
PDF Full Text Request
Related items