Text Sentiment Analysis Of Weibo Based On Deep Learning

With the rapid development of computer technology,people have entered the era of artificial intelligence.In order to further enhance the intelligence and sensitivity of the computer in the process of interaction with human beings,it can have the ability to perceive and express emotions similar to human beings.Sentiment analysis has become one of the important foundations of building a harmonious man-machine environment,and is also the future direction of artificial intelligence research.At present,microblog has become an effective channel to obtain human text emotional resources.Based on the above background,this thesis studies and designs a method for sentiment analysis of Chinese microblog comments based on deep learning.It has practical application value in computer science,social science,marketing,finance and so on.The main work is as follows:Firstly,the advantages and disadvantages of Word2 vec word vector and Bert language model are analyzed theoretically.Experiments show that Bert model is more powerful in feature extraction,feature fusion and model generalization,and can provide word vector with context semantics for downstream tasks.Then,a variable convolution kernel mechanism based on sentence length adjustment is created to enhance the generalization ability of the network model.In the past convolution network,the size of convolution kernel is often selected from the empirical value.The texts in the dataset are trained with a fixed and unified convolution kernel size,which is lack of dynamic adjustment.In order to solve this problem,we first calculate the size of the central convolution kernel according to the sentence length,and then combine it with the size of the convolution kernel nearby to form a multi-core pattern to capture emotional features.The convolution kernel size can be adjusted dynamically to avoid the simplification of features and adapt to different sentence length texts.Secondly,the dynamic routing algorithm is integrated to build the sentiment analysis network model to improve the accuracy of sentiment information analysis.On the one hand,the vector expression in capsule network is used to convey the semantic logical relationship in the text more comprehensively;On the other hand,the dynamic routing algorithm can obtain the relationship between the emotion expressed by the text fragment and the emotional tendency of the whole text,and add variable convolution kernel mechanism and maximum pooling operation to mine the hidden emotion of the text and analyze the real emotion of the whole text.Finally,the accuracy and F1 score of the network model are compared through experiments to compare the effect of sentiment analysis.The results show that the network model based on deep learning is more accurate and the test results are more suitable for users' real emotions.The research method proposed in this thesis can help the government to understand the trend of public opinion and avoid the occurrence of malignant and false events;At the commercial level,it can help businesses understand the user experience,complete the product upgrade and expand the marketing power of enterprises;At the technical level,the variable convolution kernel mechanism can be extended to other network models or text processing tasks.The overall experimental results and network model design ideas can provide clues of human emotional state for AI emotional research.