Research On The Key Technology Of Detecting And Tracking The Emergence Of Network Public Opinion

Nowadays, with the rapid development and increasing maturity of Internet technology, thepopular and widespread use of personal computer, smartphones, and other handheld terminaldevices have, more than ever, facilitated the netizens in connecting to internet, which has alsogradually deepened people‘s dependence on networks. At present, Internet has widely enteredpeople‘s work and life, and, in return, people have become more and more relied on Internet. Intimes of emergencies, more and more people are accustomed to obtain information from Internet,especially the so-called veritable truth published on microblogs, blogs and other platforms; thusInternet has become a major channel of information dissemination. At the same time, netizensalways tend to express their own emotion alongside the information dissemination. The emotionwhich is aroused by real-life events, may easily give rise to the empathy among netizens. There-fore, networks turn out to be amplifiers of public opinions. Thus, the matters concerning how tocapture emergencies from Internet, how to control the dissemination and influence of networkopinions, and how to reduce the hazard caused by emergencies, serve as a kind of statisticalsupport, having positive and realistic significance in responding to online public opinions.By far, the relative research work of emergencies in online public opinions has been deepdiscussed. The discussion on the engineering theory and method adopted in response to the so-cial science, psychology and information technology, occurrence, and development rules ofemergencies has also aroused researchers‘attention. Nevertheless, the following problems stillrequire immediate solution:(1) In the context of Internet, the fragmented dissemination of information causes dif-ficulties for netizens when obtaining all-round informationIn the context of Internet, netizens‘pursuit of uniqueness when disseminating information,together with the different distribution of netizen‘s knowledge, age and region, result in theirdisparity of focal concerns on the same matter and dissimilarity in the selected dissemination ofinformation point and volume, which finally leads to the fragmented dissemination of infor-mation. Due to the fragmented information dissemination, netizens find it rather difficult to ob-tain all-round information. Instead, the information obtained is usually the dispersed and partialfragments. The partial information also leaves room for people with ulterior motives to fabricatestories as they wish. Under certain circumstances, these information will evolve into some onlinepublic opinions.(2) Flat topic models have difficulty in describing the semantic relation between seedevents and derivative events, which causes difficulties in corresponding event detectionWhen it comes to the current detection method of corresponding events, core words aremore often applied as the basis of related events. This scheme can be applied to easily solve therelation between seed events and derivative events; yet the relation among derivative events ishard to detect. In particular, when all the derivative events are closely related to seed events, theseed features turn out to be interference in the relevance detection of derivative events. The mat- ters concerning how to remove the interference of seed events, and how to detect the relevancerelation between seed events and derivative events are the key issues to be solved in this disser-tation.(3) Current methods have ignored the impact of netizen’s emotion on the evolutionand dissemination of events, which causes the reduction of detection accuracy of emergen-ciesThe process of information dissemination is, in essence, that of emotional communicationand collision among netizens. When emergencies occur, while obtaining information from Inter-net, netizens also disseminate the information. In the process of information dissemination, theyare prone to add up their negative moods, such as fear and anger. In the process of informationdissemination, netizens find their emotion amplified and empathized, which finally gives rise tothe phenomenon of users and the stimulated outbreak of emergencies.In light of the above issues, the research content of this thesis lies in the division ofsub-topics, identification of corresponding events and detection of emergencies. The main workis given as follows:(1) To put forward a strategy of sub-topic division based on the improvement of Antcolony clustering algorithmIn this thesis, it is believed that, when a topic is divided into several sub-topics, the semanticloss of topic in the process of division should be kept to the minimum. Based on the above ideas,this thesis firstly divided the topic to the set of sentences, and constructed their feature space inthe unit of sentence; on this basis, it constructed the function of semantic loss in topic clustering,and to achieve the purpose of topic division, it improved the semantic clustering algorithm com-bined with the calculation method of sigmoid function in the construction of semantic similarity.Finally, through direct and indirect evaluation verified the feasibility of the method in this the-siss.(2) To put forward an identification method of corresponding events based on entitiesThe current research methods still draw lessons from the analytic method of relation be-tween sections and chapters. However, they fail to notice the relation between seed events andderivative events. In particular, few mentions are made to the relation inference among derivativeevents. Especially, due to the negligence of the interference caused by seed features in the identi-fication of derivative event relation, the problem arises as to the related decrease of detectionaccuracy rate of derivative events. In this light, on the basis of sub-topic division, this thesis di-vided the topic features into such two dimensions as seed features and derivative features, con-structed the model of hierarchical event on this basis, and built the clue of event relation betweendocuments from the above two dimensions; finally, based on the clues of event relation withinconstructed documents through analysis of dependency syntax, it was used as an extended clueof event relation between documents to achieve the purpose of corresponding event identifica-tion.(3) To put forward an identification method of online emergencies based on users’emotion Ever since mobile terminals‘access to Internet, the media platforms such as microblogs andblogs have embraced rapid development. Among the above platforms, however, the onlineemergencies occur more frequently than ever. Due to such features as abruptness, explosive in-crease and multiple eruptions, difficulties come along in the detection and tracking of events ofthis sorts. To solve this problem, this thesis offered a detection method of emergencies combinedwith users‘emotion. This method constructed thehierarchy model, and constantly shifted themodel‘s features driven by time sequence for the purpose of identifying emergencies online. Atthe same time, through in-depth study of users‘emotion and analysis of network users‘attitudetowards emergencies, the thesis regard the topics of negative group emotion as emergency ones,thus filtering out the large number of hot online issues concerning to life, entertainment and soon rather than non-emergencies...