A Novel Sentiment Analysis Approach Based On CPSO-LSTM

In artificial intelligence,the computational performance of Recurrent Neural Network(RNN)depends on the fine-tuning of network hyperparameters.At present,these hyperparameters are mainly set by the domain experts or by using search algorithms,where the computation cost is high.Our goal is to optimize RNN performance by using swarm intelligence to find the best combination of hyperparameters for Long Short-Term Memory networks(LSTM).In this work,we proposed a chaotic particle swarm optimization(CPSO)to optimize the hyperparameters of the LSTM model,constructed a new LSTM model,and used it for sentiment analysis of the Amazon Fine Food Reviews dataset.LSTM is a variant of RNN,which is suitable for predicting time series in different applications.CPSO is an evolutionary algorithm,and its working principle is to find the best solution through particle swarm search in high-dimensional space.In Natural Language Processing(NLP),sentiment analysis is a popular application used to mine textual opinions.We used the Amazon Fine Food Reviews Dataset,which has a total of 568,454 reviews and relates to 10 categories of food.The dataset evaluation score is 1 to 5 points.Among them,1-2 points show negative evaluations,point 3 represents neutral evaluations,and 4-5 points are positive evaluations.We used the CPSO-LSTM hybrid method with the best combination of hyperparameters to build the model,and perform sentiment analysis on the above data set,showing better classification performance.