CNN-LSTM neural network model for fine-grained negative emotion computing in emergencies

Abstract

In the Web 2.0 era, governments are facing the challenge of analyzing the emotional tendency of online public opinion during emergencies to regulate people’s emotions more effectively and maintain social stability. When dealing with large-scale short, unordered texts and extracting text features, the existing studies often face the problem of sparse features, ignoring fine-grained negative emotions. Aiming at those drawbacks and inspired by the dependency relationship among Chinese words, an emotion computing algorithm based on a binary tree is designed to assign words with emotional intensity. Then, the paper proposes a CNN-LSTM model for Chinese language sentiment classification to conduct local feature extraction and maintain long-term dependencies. The proposed model is validated using different traditional models and classifiers. The results show that the CNN-LSTM model achieved competitive classification performance. Finally, our approach was applied to practical emergency management problems, exploring the impact of government information release on negative emotion regulation to test its reliability. The experimental results validated that compared with traditional methods, this approach improved the accuracy of sentiment classification and possesses higher classification performance. The empirical analysis demonstrated that the CNN-LSTM method was rapid, effective and feasible and could be more suitable for optimizing emotion regulation policies.