Sentiment Classification Based on Integration of Rule-Based Method and Machine Learning with Image Sentiment Recognition

Abstract

Emoticons and images are often left out in traditional researches on sentiment classification. On the basis of image de-duplication, an image sentiment dictionary is constructed by correlation recognition between emoticons/images and texts. Afterwards verified image sentiment features and other rule-based features are transformed and integrated into basic feature templates. Comparison among multiple classifiers is carried out. Experimental results show that integrated features and the addition of image sentiment enhance our model’s performance, especially when dealing with text containing dialect.