Sentiment Analysis Of Web Comments Based On Label Propagation

With the advent of the "web3.0" era and the new media era,it has become an important trend for social development to use social media platforms and other carriers to obtain the latest information,convey personal opinions,and express group demands.Emotions expressed in words provide interesting and valuable information for social media services.Great value can be gained by mining the online comments containing sentiment information and making good use of them.However,a lot of manpower and material resources will be taken to complete the task of emotional analysis manually with the rapid increase of data volume.In the above situation,the technology of achieving sentiment analysis by computer automatically came along.The sentiment lexicon is the important tool for tasks of sentiment analysis,the problem of constructing sentiment lexicon is becoming one of the hot researches in natural language processing.However,the existing sentiment lexicons have problems such as limited coverage and poor domain adaptability.Building a sentiment lexicon with large coverage and strong domain adaptability has become a challenge in this field.Therefore,the semi-supervised learning method-label propagation is used to construct sentiment lexicons,and the accuracy of sentiment analysis of web comments is improved by constructing sentiment lexicons with wide coverage and strong domain adaptability.Based on the research background above,the work done in this paper for sentiment analysis of online reviews is as follows:1.In order to deal with problems such as the limited coverage and poor domain adaptability of the sentiment lexicons,a new method for selecting seed words has been proposed in this paper.Seed words are manually selected based on the general lexicon,and then the word vectors are trained on the corpus using the selected seed words manually,and finally the expanded seed words are obtained.A propagation graph and a propagation matrix will be constructed by calculating the similarity between the seed words and the candidate sentiment words,so that the label propagation algorithm is used to obtain the polarity of the sentiment words and build a sentiment lexicon.2.The algorithm of sentiment analysis based on Word2 Vec and improved mutual information has been proposed.Word vector training is performed based on corpus by selecting some hot web words manually.However,there is the problem categorizing words wrongly for Word2 Vec.Therefore,the improved PMI has been used to calculate similarity during the process of label propagation to improve the accuracy of labeling words to improve the accuracy of sentiment analysis.The effectiveness of the proposed method have been verified by experimenting on free movie review dataset and free Amazon shopping dataset in this paper.