Text Sentiment Analysis Based On Leap And Multi-attention Neural Network

With the rapid development of Internet scale,the effective processing of massive text data has become an imminent task in current society,and natural language processing tasks have emerged such as text classification and sentiment analysis.The paper is devoted to using the deep learning models to deal with the problem of sentiment analysis more effectively.By improving the current classic neural network models,it can achieve better results on the problem of sentiment analysis.1.Aiming at the problem of long training time for receiving sentence input and low efficiency when processing long texts,a leap-based LSTM-CNN model is proposed.First of all,the new model makes leap judgment when reading a text sequence,specifically,in each step,using two-layer perceptron to extract information from the front sequence,the back sequence and the current word to determine whether to skip the current word.Then,Long Short-Term Memory(LSTM)is used to analyze the sequence semantics after leap and extract its features,and the convolutional neural networks(CNN)model is used to further extract local features.Finally,it merges into smaller dimensions and outputs with positive or negative tags through softmax classifier.Compared with the LSTM-CNN model,the experimental results verify that the improved model can effectively improve the classification efficiency and accuracy.2.The combination of the LSTM network and the attention mechanism can effectively extract the text features of aspect-level sentiment analysis,but it is difficult to capture the deep sentiment feature information in the aspect information only relying on the attention mechanism.In order to better solve this problem,the paper proposes a multiattention LSTM network model.The new model first uses the attention mechanism to extract important features from the aspect information;then uses the attention mechanism to process the extracted aspect information and context information to obtain a weighted feature vector;finally,the aspect information and word vector are jointly used as the input of LSTM network.Through comparison and analysis experiments,we can know that the model can effectively improve the results of emotional analysis.This paper uses the Twitter dataset and Sem Eval dataset for experiments.The experimental results show that the leap LSTM-CNN model and the multi-attention LSTM network model can effectively improve the accuracy of sentiment analysis.