| Summary: | This study aims at constructing an Indonesian sentiment resource and improving its accuracy through the study of emotion research. This research comprises four different interconnected studies to unveil the formula of creating a good and accurate sentiment resource for Indonesian: (1) Indonesian emotion lexicon, (2) Emotion and Emotion Families in Indonesian, (3) Crosslinguistic comparison: Indonesian emotion profile, (4) Indonesian SenticNet. Here, I compiled the first Indonesian emotion lexicon created without any translation. This lexicon is equipped by the affective dimensional ratings of intensity and valence. The influencing factors of how emotion is evaluated (e.g. gender and language) were carefully observed. I also conducted the crosslinguistic comparison with other languages, especially English to highlight the Indonesian emotion profile. Despite the wide-spread claim on the universality of basic emotions, I discovered intriguing differences between the two languages. The results were then put into practice for the purpose of revamping and localizing a state-of-the-art sentiment resource SenticNet for Indonesian. In its early stage, this resource successfully achieved a satisfactory result. When tested against various datasets, it was able to predict the sentiments in a text with almost 75% of accuracy on average. The end product of Indonesian SenticNet will be mostly valuable for companies and brands that are conducting market research for their products in Indonesia. It can aid them in getting insights into their user/customer experience (UX research) and making right decisions for their marketing strategy in a faster and more accurate way.
|
|---|