Dataset of Machine Learning with Self-Assessment Manikin Valence Scale for Fine-Grained Sentiment Analysis

Sentiment analysis, the automated process of determining emotions and opinions conveyed in textual data, has attracted great interest in natural language processing and machine learning research. Traditional sentiment analysis methods typically rely on lexicons or machine learning models to classify text as positive, negative, or neutral. However, these approaches may not capture fine-grained sentiment nuance and intensity in brief, informal texts such as social media postings and reviews. Using the Self-Assessment Manikin (SAM) valence scale, we proposed a novel approach for fine-grained sentiment analysis. The scale provides a continuous measurement of sentiment that ranges from extremely positive to extremely negative. In light of this, we acquired a lexicon of emotion-laden words in Bahasa Indonesia and labeled it with a SAM valence score. At least 45 people annotated each word with a numerical score from one to nine, and the label was given for each word by averaging the scores. Furthermore, we also obtained short texts (tweets) from Twitter (X) containing words, at least one, from the lexicon. Three linguistic experts in Bahasa Indonesia annotated the texts with a numerical sentiment intensity simultaneously, also from one to nine. A fine-grained sentiment label was given for each text by averaging the intensity scores. Various methods, such as a lexicon-based polarity algorithm, (conventional) machine learning approaches, and deep neural network architectures, were developed to predict the sentiment label utilizing the labeled emotion-laden words.