Research On Chinese Emotional Classification Based On Ensemble Learning

In recent years,with the rapid development of the domestic Internet industry and the emergence of various e-commerce platforms and social platforms,the number of network users is increasing.The Internet has brought great convenience to people's lives.People can purchase goods through e-commerce platforms and evaluate the products they have purchased.These evaluation contents contain consumers' opinions and attitudes on the products,which have important reference value for the merchants to formulate marketing strategies and users to make purchasing decisions.The huge network user group has produced a huge amount of comment data.How to extract the emotional tendency expressed from the massive comment data is one of the hotspots in recent years.In order to improve the accuracy of Chinese sentiment classification,this paper proposes two kinds of Chinese sentiment classification models based on ensemble learning.:(1)An ensemble classification algorithm based on differential evolution to optimize individual classifier weights is proposed,and Chinese sentiment classification experiments are carried out on the corpus of three fields.By studying the classification methods commonly used in sentiment classification tasks,five classification models with better performance were selected.With the classification accuracy rate as the fitness value,the differential evolution algorithm is used to optimize the weight combination of the five individual classifiers,and the optimal weight combination of the five classifiers on the sentiment classification task is sought.The ensemble model is obtained through the combined strategy of weighted voting.The experimental results show that the ensemble model after optimization weights has higher accuracy in Chinese sentiment classification tasks.(2)An ensemble classification model based on Bi-LSTM is proposed for Chinese sentiment classification tasks.Firstly,word2 vec is used to train language model on corpus to get the word vector representation of all words,and use the word vector to represent the text.Secondly,using the advantage of Bi-LSTM network to learn text features more fully,multiple Bi-LSTM models are constructed.When constructing a single model,the dropout mechanism is introduced.By randomly generating different dropout retention probabilities for each Bi-LSTM model,multiple Bi-LSTM models with different network structures are trained.This parameter perturbation method enhances the diversity of individual networks in the model.Finally,the final ensemble model is obtained by combining all the models with a simple average combination strategy.Chinese sentiment classification experiments were conducted on Chinese commentary corpora in three fields,which verified the validity of the proposed model.