Towards Language-independent Sentiment Analysis

In this work, we systematically develop a Language-independent Sentiment Analysis (LISA) approach. We argue that it is generic enough to be applied across different languages/domains. Our argument is supported by an empirical evaluation showing that the proposed approach produces a competitive predictive performance if compared to others sentiment analysis approaches where there is a heavy reliance on language resources and absence of systematic pre-processing methodologies. Furthermore, when LISA is encapsulated into a multi-lingual and multi-domain version, (MLISA), we can have an accurate and compact model that can be applied to multiple languages/domains simultaneously and, hence, it suitable for online sentiment classification.