Sentiment Analysis with CNNs Built on LSTM on Tourists Comments

Abstract

This research developed a sentiment analysis system for customers' comments on a scenic spot. It is based on CNNs built on LSTM for text feature extraction under the deep learning framework. The CNNs built on LSTM model applies convolutional filters of CNNs repeatedly operate on the output matrix of LSTM to obtain robust text feature vector. In the experiments, the optimal parameter configurations for each component of CNNs and LSTM are identified separately in the first place. Then, the entire optimal parameter configuration for the integration recognition frame of the system is identified around the optimum of each component. Experimental results demonstrate that the accuracy for sentiment analysis with CNNs built on LSTM model is improved by 3.13% and 1.71% respectively, compared with a single CNNs or LSTM model.