Research On Sentiment Analysis Based On Convolution Neural Network Using Part-of-speech

In recent years,the rise of social networks has made the Internet an important platform for people to interact.Netizens have published a large number of text messages with emotional tendencies such as movie reviews and product reviews in online communities.A good understanding of user behavior has certain business research value.Most traditional sentiment analysis tasks rely on complex feature extraction and consume a lot of manpower and material resources.The rapid development of artificial intelligence in the field of deep learning has solved this problem.Using deep learning algorithms to simulate neural networks to learn things,automatically process massive amounts of text,and analyze sentiment tendencies has become a hot field of current research work.The thesis generally introduces the research status of text sentiment analysis methods in the field of deep learning,and analyzes the relevant technical characteristics of convolutional neural networks(CNN).When CNN deals with text sentiment analysis,the limitation of the size of convolution kernel will ignore the context semantic information to a certain extent,which will lead to emotional ambiguity.At the same time,it does not give different attention to the words with very emotional tendencies in the input text.In order to make up for these shortcomings,the P-CNN model and the P-Att-CNN model were proposed based on the improvement of the traditional CNN model.The P-CNN model is based on the CNN model,and uses BERT pre-trained word vectors while fusing part-of-speech features.On the one hand,it can alleviate the impact of polysemy.On the other hand,the deep two-way Transformer structure adopted by BERT can pay close attention to the semantic connection between words and words,sentences and sentences,and can achieve better semantic understanding.The P-Att-CNN model is based on the P-CNN model and introduces Attention in the P-CNN model,supplementing the CNN's advantage in capturing short-distance text features and improving the quality of model learning.In this thesis,a comparative experiment is designed for the above two improved models.The experimental results show that the model proposed in this paper obtains better classification results in the text sentiment analysis task.Compared with the traditional CNN model,the accuracy rate,recall rate,and F1 value have been significantly improved.