Analysis Of Affective Tendency For Chinese Short Texts

With the rapid development and popularization of Internet technologies,users,participation irn the development of the Internet has greatly increased.The variety of Internet social applications,such as Weibo,WeChat,and QQ,has generated an explosive growth in the amount of daily exchange data.The textual information is usually relatively short and its expression is simple.It is called short text information.In a large amount of short texts,especially users' evaluation texts on the Internet such as products,news,and people,there is a wealth of personal sentiment information.This information has a very wide range of application areas,such as online public opinion monitoring,personal emotional management,and product recommendations.Therefore,research on affective tendencies of a large number of Chinese short texts on the Internet has become a new hot spot in natural language processing research.Chinese short texts are characterized by relatively few contents,sparse data features and varied expressions.The traditional short text processing methods based on bag-ofwords paradigm has poor classification accuracy,Iow accuracy and strong dependence.In view of the above problems,this paper extracts semantically associated corpus from the relevant information such as comment forwarding of short texts to expand the original text.In terms of feature extension,in order to solve the problem of ambiguity and information sparsity,this paper first uses the keyword extraction algorithm to obtain the key generation set in the short text,then filters the retrieved Internet information and combines the extended information with the raw materials.After the fusion of features,the information consistency between the original text and the original text is ensured,and the problem of feature sparsity is solved.Then,this paper improves the single classifier by using the weight voting combination classifier and the AdaBoost based integrated learning method,and forms a combined Vote-AdaBoost classification method.Compared with the traditional method,the experimental results show that the Vote-AdaBoost combination classification method achieves 7%improvement in accuracy,recall rate and F value..Finally,based on the above work,this paper designs and implements a simple prototype system,which satisfies the functional requirements and workflow of the text classification system.The core modules of the prototype system are described in detail,and functional tests are completed.