Research On Grain Public Opinion Analysis Based On Web Mining

Grain security is related to national economic development, nation independence, andsocial stabilization in china. With the rapid development and the popular use of the Internet,the netizens can more easily public and spread all kinds of grain information. Web news havebecome one of reflecting the main carrier of grain security information. The web network hasbeen an important grain security platform of discussion and management, and plays animportant role in promoting grain security management of government. Facing anfractuousnetwork environment, it has important significance of using public opinion analysistechnology to realize effective supervision on security topics and comments, to lead thedevelopment trends of grain security public opinion correctly and in time, and to takeeffective measures to control the evolution of network public opinion. That can provide openand transparent grain security information, and has important guiding significance in theconstruction of harmonious society.Firstly, this paper analyzed the research condition of grain public opinion at home andaboard. It surveyed the key technologies involved in the web public opinion analysis, such asweb text Crawling, text representation, feature selection, weight calculation and textclassification. On this basis, this paper engages in a sophisticated discussion on the followingissues, and proposes the relevant improved mode.(1) For the shortcomings of the mutual information feature selection model, from theperspective of computing single category mutual information and relationships betweencategories, a new improved mutual information model is presented based on the weightdifferences and relationships associated with categories. It can improve the negativelycorrelation between feature words and categories, and distinguish the different degrees ofinfluence degree of feature terms. It can strengthen the feature terms relationships associatedwith categories, and also adjust and optimize feature weight. The effectiveness of theimproved algorithm is verified by experiments.(2) As a common weighting algorithm, TFIDF model does not make use of thedistribution of feature term in documents and take into account the distribution of feature terms in categories. These lead not to choose the better terms to represent documents content.This paper proposes a new TFIDF feature weighting method based on position importanceand entropy factor. Position importance could strengthen the content structure of the newsdocuments, and entropy factor can reflects distribution of text documents including featureterms among categories. Then the experiments prove that, the proposed method achieves abetter classification results.Based on the researches, this paper developed a grain public opinion analysis demosystem, and designed the function modules such as grain information acquirement, graininformation preprocessing, and grain public opinion analysis and so on. Then this paperimplements the function of the topic tracking and information summary of network publicopinion.