Study And Implementation On Opinion Shift Detection Technology For Web Hot Event

With the development of Internet technology, network has gradually become the place to express their opinions, ideas and emotions for people. Network public opinion is receiving more and more attention, and researchers now pay more and more attention on the research about network public opinion information. At present, the researches about network public opinion are mainly concentrated on information collection, classification and summary about microblogs, blogs and comments, but the research about the comments of network hot events is less. So in this thesis, from the subjective texts such as comments of network hot events, the opinion content drift detection and the emotional polarity drift detection about network hot events are focused on.For the opinion content drift detection about network hot events, according to the characteristics of the network hot events, latent semantic of a single day’s comments is extracted with LDA model and applied for modeling and clustering. Then the key clusters in a specified period from the clusters obtained are recognized, and a key words set in the key clusters is extracted, which are on behalf of the central events. Finally, for the opinion content drift detection, the semantic similarity between key clusters is calculated by the semantic similarity between words of them, and then according to the semantic similarity of key clusters in the adjacent time, whether opinion content drift happened is tested.For the opinion emotional polarity drift detection about network hot events, according to the characteristics of the comments set about the network hot events, the comment data set over time is divided to get the comment subsets sequences. Then each sentence is regarded as a unit. Firstly, the emotional polarity value of every sentence is calculated. Then the emotional polarity value of every comment is calculated by the ones of all sentences in the comment. At last, the emotional polarity value of every subset is determined by the ones of all comments in the subset. Finally, according to the different time zone emotional polarity distribution, whether emotional polarity drift happened is tested.The experiments show that the proposed method can effectively test opinion content drift and opinion emotional polarity drift of network hot event comments, and get a good result in the two detections.