Research On Chinese-Vietnamese Cross-language Object-level Sentiment Analysis Method For Social Media Text

With the rapid development of the Internet,the level of mastery of public opinion information in social media data needs to be gradually improved.The Chinese-Vietnamese cross-lingual object-level sentiment analysis captures and analyzes the sentiment tendency of Chinese and Vietnamese attention seekers under the same event,and grasps the public opinion dynamics of both countries to further carry out analysis,monitoring and early warning tasks about hot events.The Chinese-Vietnamese cross-lingual object-level sentiment analysis for social media texts is a domain-specific task with research problems such as lack of annotated data,inaccurate recognition results,inaccurate sentiment representation mapping,and insufficient learning of comment features,which deserve in-depth research.In this thesis,we study the Chinese-Vietnamese cross-lingual object-level sentiment analysis method for social media texts,mainly from the following aspects:(1)Chinese-Vietnamese cross-linguistic sentiment analysis corpus construction:the construction of datasets with annotation is the basis of supervised learning,and this thesis proposes a dataset construction method for social media comment opinion object analysis.First,social media comment data related to hot events are collected from Sina Weibo and Twitter platforms.Secondly,a Chinese-Vietnamese cross-lingual viewpoint object classification dataset and a Chinese-Vietnamese cross-lingual sentiment classification dataset are constructed according to the different requirements of the Chinese-Vietnamese cross-lingual viewpoint object recognition task and the Chinese-Vietnamese cross-lingual sentiment classification task,which lay an important foundation for the later research.(2)A graph neural network-based approach to Chinese-Vietnamese cross-lingual viewpoint object recognition: viewpoint object recognition on Vietnamese comments is accomplished through Chinese viewpoint object labels and the corresponding alignment knowledge.However,labeled data are scarce,cross-lingual comments are complexly associated,and modeling comment representation is difficult.Considering that the same viewpoint objects exist between Chinese and Vietnamese comments when discussing the same events.Therefore,this thesis proposes a graph neural network-based method for the recognition of cross-lingual viewpoint objects between Chinese and Vietnamese.By constructing a heterogeneous graph including Chinese-Vietnamese comments and keywords,the complex association relationships in Chinese-Vietnamese comments are effectively modeled,and the graph structure is used to achieve neighborhood information aggregation and comment node update.Experimental results on the Chinese-Vietnamese cross-lingual viewpoint object classification dataset show that the method in this thesis is more effective and advanced compared with the benchmark method.(3)Chinese-Vietnamese cross-lingual sentiment classification model incorporating opinion features: by transferring Chinese sentiment knowledge to Vietnamese comments,the sentiment classification task of Vietnamese comments is completed.While the existing models are difficult to solve the problems of insufficient learning of sentiment representations and accurate mapping of cross-lingual sentiment representations into the same feature space,considering the role of viewpoint object information in enhancing the learning of sentiment representations and reducing linguistic differences,a Chinese-Vietnamese cross-lingual comment sentiment classification model incorporating viewpoint object features is proposed to fuse and encode viewpoint object representations with semantic representations through a gating mechanism,and use adversarial learning The model learns the representation with the smallest difference in language distribution,and finally trains the model classifier to complete the sentiment classification task by using Chinese comment labels.The experimental results show that the model in this thesis can fit the differences in language distribution faster and get richer sentiment representations,and the experimental results are significantly improved compared with the baseline model.(4)Prototype system for analyzing the viewpoint objects of Chinese and Vietnamese social media comments: Based on the above-mentioned theoretical research,this thesis designs and implements Chinese-Vietnamese social media comment viewpoint object analysis prototype system,which is developed and built with Vue framework,designed by element-plus,the system uses route forwarding technology to convert the model into an interface integrated into the system,which integrates the functions of Chinese-Vietnamese social The system integrates the functions of Hanover social media comment data collection,Hanover social media comment opinion object recognition and Hanover social media comment sentiment classification.The system also provides a visual interface to facilitate quicker and more convenient access to the service.