Research On Recommendation Of Network Public Opinion Hotspot Keywords Based On Clustering

With the development of network information technology and people's diverse demand, Internet has become a primary public opinion channels for expression of interest, emotional catharsis, thoughts collision. Therefore, it is of great significance for deeply research on Internet public opinion mining. In particular, public opinion monitoring refers to the integration of Internet information collecting and intelligent information processing technology, through the Internet mass information acquirement, automatic classification, clustering, topics monitoring, themes spotlight, realizing the customer's demand of public opinion monitoring and tracking. In recent years, public opinion system meet customers'requirement, not only on the information monitoring, but creating a lot of applications, they provide analysis basis for customers in grasping information comprehensively and making the right public opinion guidance.After we summarize and analyze the current development trend of public opinion system and the recommend system which is widely used in our Internet, in this paper, considering the characteristics of public opinion system, we break the traditional method for crawling information that users configure monitoring keywords by their own ideas, but initiatively recommending some monitoring keywords for users according to their behavior.Before our recommendation, we should authorize some operations for customers, when a document was rated by users. We call them high-value documents based on these documents. First we conduct words segmentation by using HMM and Viterbi algorithm, then using TextRank to extract keywords, thus we can representation these documents by keywords, then using k-means++ algorithm to clustering them and recalculating keywords'weight, finally we can obtain the recommended keywords from every cluster. Through this way, users can not only obtain some new choices, but get some hints. It is not simply contains the original keywords or combination, but get the keywords from the documents which they think it is very important.