KASN: Knowledge-Aware Siamese Network for sentiment analysis

Sentiment analysis is essential in natural language processing (NLP) which aims to extract sentimental opinions or polarities in given texts. Deep learning-based sentiment analysis models usually demand a large amount of training data. In some applications, however, only limited training data is available, which in turn deteriorates the performance of deep learning models. To address this problem, we propose a novel approach that effectively incorporates external knowledge into deep learn- ing neural networks by comprising a siamese network-based similarity learning module for target and pseudo sentences. Our model can work with many existing deep learning models and improve their performance in a low-cost fashion. We evaluate our model on three popular datasets in sentiment analysis. Experimental results show that our model outperforms major baseline deep learning models.