Research On Sentiment Classification Of Weibo Based On Word Vector Emotion Enhancement And Multi-channel Model

Weibo is one of the most active social platforms on the Internet in China.Emotional classification of Weibo comments is of great significance in the areas of public opinion control,business decision-making and counter-terrorism.At present,there are relatively few researches on sentiment classification of Weibo reviews.There are serious text colloquialization,high frequency of homophones,and the emergence of new types of network languages.The problem that models are difficult to accurately learn the emotional information of text due to the traditional word vectors' insufficiency in representing emotional information also needs to be solved urgently.In view of the fact that the current sentiment dictionary is mainly based on the standardized language and ignores the non-standard languages such as network language,this paper statistically sorts out a list of common network languages,and uses the open source sentiment dictionary and Weibo comment text to build a network language sentiment dictionary to improve the accuracy of language segmentation,and make full use of the emotional information included in it.In view of the lack of emotional information in traditional word vectors,drawing on the psychology of Prucheck's emotional roulette theory to construct emotional vectors,this paper proposes a multi-emotional word vector refining algorithm.The algorithm adjusts the pre-trained word vectors from the perspective of multiple sentiment through the emotion vectors,to enhance the sentiment representation of word vector and avoid the loss of semantic information of word vector.Experiments show that the emotion sentiment word vectors can effectively improve the emotion classification performance of multiple models.In view of the characteristics of microblog text and the relatively single feature extraction of existing models,this paper uses the characteristics of convolutional neural networks and recurrent neural networks to extract features at the three levels of words,parts of speech and pinyin,to build a multi-channel Bi-LSTM-CNN sentiment multiclassification model,and this paper constructs a word-POS-senti attention mechanism by introducing part-of-speech and sentiment vectors into the attention mechanism.Experiments show that the classification model has better comprehensive performance than other models,and the word-POS-senti attention mechanism can effectively improve the convergence speed and classification performance of the model.