Automatically generating a sentiment lexicon for the Malay language

Abstract

This paper aims to propose an automated sentiment lexicon generation model specifically designed for the Malay language. Lexicon-based Sentiment Analysis (SA) models make use of a sentiment lexicon for SA tasks, which is a linguistic resource that comprises a priori information about the sentiment properties of words. A sentiment lexicon is an indispensable resource for SA tasks. This is evident in the emergence of a large volume of research focused on the development of sentiment lexicon generation algorithms. This is not the case for low-resource languages such as Malay, for which there is a lack of research focused on this particular area. This has brought up the motivation to propose a sentiment lexicon generation algorithm for this language. WordNet Bahasa was first mapped onto the English WordNet to construct a multilingual word network. A seed set of prototypical positive and negative terms was then automatically expanded by recursively adding terms linked via WordNet’s synonymy and antonymy semantic relations. The underlying intuition is that the sentiment properties of newly added terms via these relations are preserved. A supervised classifier was employed for the word-polarity tagging task, with textual representations of the expanded seed set as features. Evaluation of the model against the General Inquirer lexicon as a benchmark demonstrates that it performs with reasonable accuracy. This paper aims to provide a foundation for further research for the Malay language in this area.