The Modeling And Implementation Of Text Sentiment Classification Based On Skip Method Model

With the rapid development of social networks,more and more people are accustomed to express their opinions,share their lives and obtain information they are interested in through social platforms.By analyzing the emotional tendency of the text,we can not only know the personal preferences of users,but also monitor public opinions based on the information.Text sentiment classification is one of the important tasks in natural language processing tasks.Recently,the neural network model has achieved good results in natural language processing tasks.This paper proposes some improved neural network models for emotional classification tasks,which have good performance.The main work of this paper is as follows:Firstly,considering that the input data of the model is not all related to the task,it is proposed to add skip mechanism into the model,so that the model can self-adaptively skip the data which is unrelated to the task.In order to implement the skip mechanism,two methods are proposed in this paper.The Skip-Before mechanism means that before the model calculates the hidden state of the current data,it decides whether to skip the current word according to historical information.If the word is to be skipped,the model no longer calculates the hidden state of the current word.The Skip-After mechanism means that first calculates the hidden state of the current input,and then decides whether to retain the calculated hidden state according to the historical information.If the word is to be skipped,the calculated hidden state value is discarded.Secondly,the two skip mechanisms are applied to the RNN model and the LSTM model,and four models are constructed namely Skip-Before-RNN,Skip-Before-LSTM,Skip-After-RNN and Skip-After-LSTM.These models were applied to adding-task,using simulation data to verify the model capability,at first the results are not effective,so that a method of using dynamic learning rate is proposed to improve the models.The results of experiment verify the feasibility of the model.Thirdly,these models were used in text sentiment classification,and a series of exploratory experiments were conducted using the IMDB film review dataset and the Stanford Sentiment Treebank dataset.The experimental results prove the effectiveness of the skip mechanism proposed in this study.The model can adaptively learn the skip behavior and skip the words in the text that are not related to the task.To some extent,the computational cost of the model can be reduced and the model can be improved.Also the experimental results show that the RNN model based on skip mechanism is of poor robustness,with good effect in some cases and poor effect in others.The LSTM model based on skip mechanism can effectively skip the irrelevant words in the text and improve the training effect of the model.At last,two methods were tried to optimize the model.The methods and key points of optimizing the experimental process and improving the model training process are summarized,which provide reference value for later researchers.The innovations of this paper are as follows:(1)The Skip-Before mechanism is proposed.The main idea is that the neurons in the model first calculate the skip probability of the input,and determine whether to perform the operation of the hidden state according to the probability value.The SkipBefore-RNN model and the Skip-Before-LSTM model based on this mechanism are implemented in this research.(2)The Skip-After mechanism is proposed.The main idea is that the neurons in the model first calculate the current hidden state value,then calculate the skip probability of the input,and determine whether to perform the update operation according to the probability value.The Skip-After-RNN model and the Skip-AfterLSTM model based on this mechanism are implemented,and the simulation results of the two models are used to verify the feasibility of the model.These two models are used in the text sentiment classification task.(3)In the construction of Skip-Before-RNN,Skip-Before-LSTM,Skip-AfterRNN and Skip-After-LSTM models,in order to realize the model-adaptive learning of skip behavior,the dynamic learning rate method is used to calculate the skip probability to improve the model.