A Robust Sentiment Analysis Method Based on Sequential Combination of Convolutional and Recursive Neural Networks

With explosive development of the World Wide Web, an enormous amount of text information containing users' feeling, emotions and opinions has been generated and is increasingly employed by individuals and companies for making decisions. Whereas unstructured form of data must be analyzed to extract and summarize the opinions in them, sentiment analysis has changed to a significant research area in the field of Natural Language Processing. In this regard, deep learning methods have attracted a lot of attentions in recent years and various deep learning models have been proven as effective network architectures for the task of sentiment analysis. However, each of them has its potentials and weak points. To eliminate their drawbacks and make optimal use of their benefits, convolutional and recursive neural network are merged into a new robust model in this paper. The proposed model employs recursive neural network due to its tree structure as a substitute of pooling layer in the convolutional network with the aim of capturing long-term dependencies and reducing the loss of local information. The proposed model is validated on Stanford Sentiment Treebank by conducting a series of experiments and empirical results revealed that our model outperforms basic convolutional and recursive neural networks while requires fewer parameters.