Research On Topic Discovery And Emotional Analysis Of Micro-blog Public Opinion In Bursty Events

With the rapid progress and development of electronic technology and Internet technology in the world and China,as well as the popularity of modern intelligent handheld communication devices,Sina Weibo,because of its low threshold and free information dissemination characteristics,has become a platform for modern people to access information and share their lives.When users want to express their views or opinions,they can be published through Weibo.They can also retweet,follow and comment on other users' microblogs.Therefore,Weibo has become an important tool and carrier for the dissemination of public opinion in bursty events,and plays an important role in the dissemination of public opinion.How to accurately find out the opinions and topics of public opinion caused by the bursty events and analyze people's emotions has become an important hot issue for social experts and scholars to study.Therefore,it is of great reference value and guiding significance for enterprises and governments to timely dig out public opinion topics and grasp public sentiment from massive microblog data.This thesis focuses on the public opinion of Weibo on unexpected events,and on topic discovery and fine-grained micro-blogging sentiment analysis as the research content.By reading a large number of relevant literature,this thesis analyzes and summarizes the research status of topic discovery and emotion analysis,organizes the relevant materials and combs the basic theoretical knowledge.On this basis,this thesis conducts an in-depth study of topic discovery and fine-grained Weibo emotion analysis.The main research contents of this thesis are as follows:1.Topic discovery of Weibo public opinion.Due to the concise language of the Weibo text and the small amount of characters,the traditional LDA topic model is not effective for short text Weibo topic discovery.Aiming at the problem of short text content of some microblogs,which affects the effect of topic discovery,this thesis proposes a short text expansion strategy.On this basis,a topic discovery model based on the LDA model and BC-BIRCH hierarchical clustering algorithm is proposed.This model combines the same topic clusters to improve the effect of topic discovery.Compared with a single topic discovery model and clustering algorithm,the model proposed in this thesis has certain advantages in topic discovery effect,and is suitable for topic discovery of microblog public opinion.2.Fine-grained sentiment analysis of Weibo.Aiming at most of the current sentiment classification models still staying on flat and coarse-grained classification problems,this thesis proposes a multi-level microblog fine-grained sentiment analysis model.This model combines Naive Bayes and an improved stochastic gradient descent method.For the coarse-grained sentiment classification of Weibo,this thesis adopts Naive Bayes classifier.For fine-grained sentiment classification,an improved stochastic gradient descent classifier is used.The overall idea of the model is to first achieve the local optimal,so as to achieve the overall optimal.Compared with the planar classifier,it not only greatly reduces the computational complexity and improves the classification efficiency,but also improves the classification effect obviously.In this thesis,the real data of Sina Weibo is collected,and the above working methods are tested and verified.Through data comparative analysis,the experiment is compared with the topic discovery model and fine-grained emotion analysis method proposed in this thesis.The experimental results show that the proposed model and method are better than other algorithms.