Research On Topic Extraction Method Based On Sentiment Analysis And Feature Screening

In recent years,the Internet has penetrated into our daily life,and APP has become one of the tools for life.When people buy goods,they often look at the reviews of the products to make purchase judgments on the goods.These text comments are the true opinions of consumers,and also provide a way for the platform to understand the customers.So the problem should be solved as soon as possible,which quickly mine information in text.This paper explores the methods of sentiment analysis and topic mining which based on the APP short text data.This paper briefly introduces the text preprocessing technology,text feature extraction technology and unbalanced data processing technology.The main task is to explore the quick and accurate text sentiment analysis method and the topic extraction model.The sentiment analysis and feature screening are integrated into the topic model,and a topic extraction method based on LDA model is proposed.The main research contents of this paper is as follows:My job is to explore and select the unbalanced data technology suitable for the short comment data.The resampling and undersampling technology were used to process and compare the results.The sentiment analysis task includes the following two aspects: the sentiment analysis based on machine learning model and the sentiment analysis based on deep learning,to look for models for unbalanced data.The machine learning model are support vector machine and Extreme Gradient Boosting,with Grid Search and Cross-validation to adjust parameters.Deep learning model is fastText.Three models are evaluated by F1-Measure.The resultds show that the fastText is better than the other two models in dealing with unbalanced data,and the reason is pointed out.In the topic extraction task,a method based on LDA model for short and unbalanced text data is proposed.Firstly,sentiment analysis is carried out.The label of emotional polarity is added to the topic extraction model,and then feature screening is performed to delete public attribute features.The short text topic extraction is more effective than the original model.Thus the method has practical application value.