M-COVIDLex: The Construction of a Domain-Specific Mixed Code Sentiment Lexicon

Abstract

Sentiment lexicons serve as essential components in lexicon-based sentiment analysis models. Research on sentiment analysis based on the Malay lexicon indicates that most existing sentiment lexicons for this language are developed from official text corpora, general domain social media text corpora, or domain-specific social media text corpora. Nonetheless, none of the current sentiment lexicons adequately complement the corpus utilized in this study. The rationale is that words in established sentiment lexicons may convey different sentiments compared to those in this paper’s corpus, as the strength and sentiment of words are context-dependent, influenced by varying terminology or jargon across domains, and words may not share the same sentiment across multiple domains. This paper proposes the construction of a domain-specific mixed-code sentiment lexicon, termed M-COVIDLex, through the integration of corpus-based and dictionary-based techniques, utilizing seven Malay part-of-speech tags, and enhancing Malay part-of-speech tagging for social media text by introducing a new tag: FOR-POS. The constructed M-COVIDLex is evaluated using two distinct domains of social media text corpus: the specific domain and the general domain. The performance indicates that M-COVIDLex is more appropriate as a sentiment lexicon for analyzing sentiment in a domain-specific social media text corpus, providing valuable insights to governments in assessing the sentiment level regarding the analyzed topic.