Text Sentiment Analysis Based On Attention Mechanism And BiGru

With the rapid development of network technology and the Internet use cost is reduced,Internet users are increasing year by year,more and more people tend to be on the network platform to comment,like a commodity,such as large hot topic review with emotional color of text information,analyze the subjective text can better understand the user's emotional attitude,has a certain commercial applications.How to extract valuable information from massive unstructured texts has become one of the research hotspots.Sentiment analysis,as a sub-task of natural language processing,aims to analyze people's emotional tendencies and opinions about entities and their attributes from text.Traditional coarse-grained sentiment analysis can only analyze the emotion of the whole sentence,but cannot identify the emotional polarity of specific aspects in the text,so the granularity of sentiment analysis needs to be more refined.Based on the attention mechanism and the Gate Recurrent Unit(GRU),this paper conducts a textual sentiment analysis at the aspect level.The research work and main innovation points of this paper are as follows:To solve the problem of neglecting aspect word information in previous studies,a two-way GRU aspect level sentiment analysis model(ATAE-BiGRU)based on attention was proposed.The model uses the BERT pre-training model to mine the semantic association and its feature representation of the text.The target and context are modeled through the GRU network to obtain the hidden vector representation of the target and context,and the semantic information is extracted.The attention mechanism is used to capture the key information in the sentence and assign weight scores to different words.The hidden state is combined with aspect word vector.The aspect word vector is trained as the model parameter to obtain the weight representation of the text on the given aspect word.In this paper,we conduct experiments on public data sets,and the experimental results show that the two-way GRU model based on attention is superior to the traditional neural network model.In aspect level text sentiment analysis,the traditional neural network model has little interaction between context and aspect words,cannot fully consider the grammatical structure information of words,and ignores the influence of location information on the emotional polarity of aspect words.This paper proposes an attentional interaction model with location information(MAIP).The model uses an unsupervised neural language model,Glove to train the initial word embedding.Dependency analysis tree is introduced to represent the correlation information between words.The position information is embedded into the word vector.Analyze the degree of association between aspect words and context.The experimental results show that the proposed MAIP model achieves good classification performance in the text sentiment analysis task.