Sentiment Analysis(SA)is a very important sub-task of Natural Language Processing(NLP)in the Internet era,which can help users to analyze comments and public opinion.However,most existing studies focus on improving the overall performance of sentiment analysis tasks,and there are few categorical studies targeting different text features.Classification research can help researchers find the shortcomings of current analysis methods in specific scenarios,and can also guide users to choose more appropriate analysis methods when facing different scenarios.Based on the BERT model,SentiHood is classified according to text length and number of evaluation targets,and different analysis methods are used to conduct group experiments.The experimental results show that each method performs better in short text analysis than in long text analysis,and the performance of single objective analysis is better than that of multi-objective analysis.Under different text features,there are different analysis methods that demonstrate optimal performance.
国家科技期刊平台
NSTL
万方数据
