Research On Text Sentiment Analysis Based On Fusion Neural Network Model

Under the influence of economic globalization and the iterative development of information technology,the Internet business has grown speedily,in recent years,the data information on the network has been constantly proliferating,and the forms are increasingly rich,more and more people tend to use social platforms to express their subjective views and comments.By sorting out massive unstructured text data,digging into valuable information and classifying subjective comments by users,we can not only better understand the needs and preferences of most users,but also help the regulatory authorities to regulate online public opinion and prevent public opinion from getting out of control and causing immeasurable harm to social stability.In view of the limitations of Convolutional Neural Network(CNN)and Recurrent Neural Network(RNN)in text feature abstraction,a deep learning model of multi-channel convolution and Bidirectional Gated Recurrent Unit(BiGRU),Pt-MCBGA,is put forward in this paper,which references an attention mechanism and a pre-trained word vector model.So that it can better extract the feature information in the text,mine the sentiment polarity in the text,and analyze the sentiment trend of the topic participants.The model abandoned the traditional way of building an index dictionary and training word vectors with Word2 vector model,and used the pre-trained word vector model to directly fill in the high-dimensional word vector,which avoided the problem of insufficient training and inaccurate representation of word vector due to the lack of data text.Using the multi-channel one-dimensional convolutional neural network Conv1D and BiGRU to mine text information,more local features and global semantic features of different granularities can be extracted.By introducing the attention mechanism and weighting the features,the model can quickly obtain more key feature information and improve the possibility of outputting the correct probability distribution.Through comparative tests,we find that using accuracy,precision,recall and F1-score as assessment norm,compared with other model evaluation index system values,the model has a certain degree of improvement.This paper shows the applicability of the Pt-MCBGA model in sentiment text classification in the form of actual cases,and verifies its applicability compared with other deep learning algorithms in the task of sentiment text classification.The algorithm has strong practical significance and application value.