Research On Positive And Negative Comments Classification Algorithm Based On Attentional Mechanism

During the big data times,users have a variety of evaluations on network software platforms.These evaluations can directly reflect users'views on an event or product.Emotional analysis of these positive and negative comments will provide reliable data support for market monitoring and research,user recommendation,the preparation and survey in advance of popular feelings on the network.Target at complexly positive and negative evaluation text,the effect of traditional emotion analysis feature extraction method is not significant.This paper makes the multi-level Attention mechanism into the emotion classification of microblog comment data,and combines the BiLSTM Algorithms model.Finally realizing an improved emotion classification algorithm model and named it as the Multi-level Attention-BiLSTM Sentiment Classification Model.The experimental results show that the accuracy of the model can reach 91.8%in the text classification task of microblog positive and negative comments.Compared with the Single-layer Attention-BiLSTM algorithm and BiLSTM algorithm,the accuracy of the model is improved by 0.002 and 0.008 respectively.On this basis,using the idea of ensemble learning to fuse a variety of strong classifiers,finally the accuracy of micro blog classification task reaches 92%,which is improved by 0.002.The main work of this paper is as follows:1.Text acquisition and preprocessing.Collect and arrangement 120000 positive and negative comments on microblog,uses Python's genism library to train model and then trains word vector through FastText text representation model,uses TF-IDF of each word as weight,and sets appropriate length limit combined with sentence length of text set,and carries out zero filling and truncation operation on text.2.The emotion classification algorithm model of Multi-level Attention-BiLSTM is implemented and optimized.This paper combines the attention mechanismand BiLSTM long-term and short-term memory neural network model.Finally,it was named as the Multi-level Attention-BiLSTM Sentiment Classification Model.To design a two-layer attention mechanism to train the weight of words and sentences respectively,and optimize the model by adjusting the super parameters.Designing several groups of comparative experiments to prove the Multi-level Attention-BiLSTM algorithm can improve the classification effect.3.Research on integrated learning and fusion of multiple algorithm models.For reasons of the less accuracy and bad generalization performance in the emotion classification task,by adopting the idea of integrated learning model integration,combine these five classifiers of the Multi-level Attention-BiLSTM algorithm model,XGBoost,Bayes,LSTM,BiLSTM by bagging way.And fusing them into a more robust classifier model,and applied to the microblog comment text emotional category.At the last,by the contrast test demonstrates that the functionality and superiority of fusion.