A Multi-channel CNN-BiGRU Model For Chinese Microblog Sentiment Analysis By Fusing Word Embedding And Attention Mechanism

With the rapid development of Internet technology,Internet social media has become an increasingly important platform for emotional communication in human life.Sina Weibo has become one of the most popular social media in China due to its massive user base,wide reach and fast sharing of real-time information.Collecting a large number of emotional comments on Weibo and analyzing them provides powerful data for monitoring public opinion and selling products.Convolutional neural network(CNN)and recurrent neural network(RNN)are widely used in the field of text sentiment analysis,and have achieved good results.However,there is a problem of context dependence among texts,and CNN ignore the semantic information of the context between words when extracting the local information among consecutive words in a sentence.Bi-directional gated recurrent unit(BiGRU)network can not only solve the gradient disappearance or gradient explosion problem of traditional RNN models,but also well compensate the disadvantage that CNN cannot effectively extract the contextual semantic information of long text.To address the problems of inadequate feature extraction from a single word vector and lack of focus on key information,this paper proposes a multi-channel CNN-BiGRU model that fuses word embedding and attention mechanisms,which fuses word vectors generated by three word embedding models,Word2Vec,FastText and Doc2Vec,at the embedding layer to fuse semantic and sentiment information and capture inter-word,The multi-channel computational structure integrates the advantages of convolutional neural networks and bidirectional gated recurrent units in extracting text corpora,and introduces an attention mechanism to further improve the model’s ability to focus on key features.In this paper,we experimentally validated the Sina Weibo dataset where netizens discussed around the topic of the epidemic during the New Crown epidemic.The experiments showed that the proposed model improved the accuracy of CNN and BiGRU by 5.89%and 4.56%,and the F1 values by 5.32%and 4.94%,respectively,on the Chinese dataset compared with the baseline model.