Research On Chinese Implicit Sentiment Analysis For Social Media Texts

With the popularity of Internet technology in China,Internet social media has reached in a blowout growth.A large amount of textual data implied with sentiments are produced while the native netizens use the social media platform to communicate with others.According to the expression whether there are certain sentiment words,researchers divide them into explicit sentiment expression and implicit sentiment expression.There are so many of the native netizens accustomed to expressing themselves implicitly that the quantity of Chinese implicit sentiment expression is huge.Therefore,the analysis of such large amount of implicit sentiment expression is vital in the research area of sentiment analysis,which has broaden application scenarios and important research significance in public opinion analysis,user experience improvement and service and product perfection.After analyzing the text data from social media platform,this paper defines two major problems in Chinese implicit sentiment expression:(1)the problem of "weak features" without obvious sentiment features in sentiment expression;(2)the problem of "multi-confounding weak features" with multiple sentiment orientations and no obvious sentiment features in sentiment expression.This passage dives into the methods of Chinese implicit sentiment analysis oriented towards social media text,which aims to tackle down the issue of "weak features" and " multi-confounding weak features " raised by the Chinese implicit sentiment expression based on the textual data originated from the social media.The main research work of this paper is as follows.(1)Proposed ’’Hierarchical Knowledge Enhancement and Multi-pooling’’(HKEM).The model uses a hierarchical knowledge enhancement method to fully integrate and learn different levels of knowledge information in the text to alleviate the "weak features" problem.At meanwhile,we model the problem of Multiple Confusion of Weak Feature with multiple features of learnt text extracted and analyzed by using the multi-pooling method.In this paper,comparative experiments have been conducted on the SMP-ECISA2019 dataset,and the results how that the model proposed by this passage makes no distinction of rank to the best model while the F1 score output by the former outperforms the latter one about 5.9% which makes the former one reach into the state-of-the-art model,which proves the effectiveness and superiority of the model proposed in this chapter.(2)Proposed "Fusion Sememe Knowledge Representation"(SKR),which is to alleviate the problems of "weak features" and "multi-confounding weak features" from a perspective different point from the previous research by using external sememe knowledge.This model integrates the sememe knowledge in the process of learning the text,which can effectively describe and depict the text semantics in detail,improve the training quality of low-frequency words in Chinese implicit sentiment expression,At the same time,on the basis of learning the semantic relevance of words,modeling work closer to the nature of the text can be carried out to enhance the understanding of the text.According to the experiments,the proposed model outperforms those contrast experiments in the term of F1 score,which means that it has greater ability of analyzing the Chinese implicit sentiment expression and relieving the problem of "weak features" and "multiple confusion weak features" in some extent.Therefore,the efficiency and effectiveness of proposed model are well demonstrated.(3)Build a "Chinese Implicit Sentiment Analysis Prototype System".The system integrates the Chinese implicit sentiment analysis model proposed in this paper.Firstly,it realizes the reception of the input Chinese text and transmits it to the system backend.Besides,the probability of three kinds of sentiment tangency is calculated and the results of sentiment analysis are output by the model.Finally,we visualize the above results to the users.