Improving accuracy in sentiment analysis for malay language

Accurately determining the intent of a person from their writings is one of the key challenges in developing Sentiment Analysis methods and tools. Among the various proposed methodologies in the literature, Machine Learning based approaches are widely used, especially for multi-lingual text content. This paper presents a hybrid approach of a Knowledge Base approach combined with Machine Learning, implemented in Mi-Intelligence Sentiment Analysis system. The proposed approach supports multilingualism and is applied on text articles written in Malay (Bahasa Melayu) language. A dataset in Malay language is manually annotated with sentiment values and used for performance evaluation. The results are compared with other Machine Learning tools and analysed. The proposed approach achieved an accuracy of 94.34% and outperformed the traditional pattern matching approaches by being able to take into consideration the meaning of the text.