Text Classification In Bidirectional GRU Network Based On Attention Mechanism

With the explosive development of Internet,network information fills our life,and these information contents are text,video,image,speech and other types,among which text information,as the most basic data,is also easier to obtain from the network,but these text data often have the problem of messy and disorganized sentiment information,so the sentiment classification in text classification research is to solve this problem.Extracting features from a large amount of text data often results in problems such as duplicated information,inconspicuous features,and large feature dimensions,which lead to unsatisfactory sentiment classification results.Therefore,in order to quickly and accurately locate the emotionally inclined keywords we need,the study of text sentiment classification in natural language processing has become a hot topic.The main research work on sentiment classification problem in this paper has two aspects as follows:(1)To address the problem that key information cannot be accurately extracted from Chinese text,resulting in low accuracy of text sentiment classification,this paper proposes to introduce the Viterbi algorithm in the word vector training model to calculate the word vector probability with the highest probability in the word vector training process,i.e.,to find the word vector probability that best matches the semantic words in the current context,and then input to the Bi GRU-Attention model for The extracted feature information is given different weights and finally the processed data is classified in the Softmax classifier.The experimental results show that the Viterbi algorithm introduced in the word vector training model proposed in this paper improves the accuracy of text sentiment classification.(2)The existing sentiment analysis models have problems such as feature duplication and inconspicuous features in the feature extraction process,and the CBOW and TF-IDF fusion models are proposed for word vector training.Firstly,the input data are trained using both the CBOW model combined with the Viterbi algorithm and the TF-IDF model,and the two sets of word vector matrices obtained are averaged and then input to the Bi GRU layer and the Attention layer,and the information extracted by the model is given the corresponding weights and the key information is highlighted.Finally,the processed data are classified in the Softmax classifier.The experimental results show that the word vector tolerance model proposed in this paper not only improves in feature extraction,but also improves the classification effect of the model.