Sentiment Analysis in News Articles Using Sentic Computing

Fine-grained sentiment analysis in news articles is a challenging problem with many potential applications. The difficulties of performing sentiment analysis in this domain can be overcome by leveraging on common-sense knowledge bases. In this paper, we present an opinion-mining engine that exploits common-sense knowledge extracted from ConceptNet and SenticNet to perform sentiment analysis in news articles. We have tested our engine on a large corpus of sentences from news articles. Our results show 71% accuracy in classification, with 91% precision for neutral sentences and F-measures 59%, 66% and 79% for positive, negative and neutral sentences, respectively. Our method can potentially be applied to reputation management in text-based media such as newspapers.