Impact of Statistical and Semantic Features Extraction for Emotion Detection on Indonesian Short Text Sentences

The ability to detect emotions in short texts is crucial because interpreting emotions on platforms like Twitter can offer insight into social trends and responses to specific events. Additionally, examining emotions in product reviews assists companies in comprehending customer sentiment, allowing them to improve the quality of their products and services. Most research on Indonesian language emotion detection utilizes statistical feature extraction, with limited discussion on the impact of both statistical and semantic feature extraction. Thus, the research aims to detect emotions in short texts equipped with an analysis of the impact of statistical and semantic features. Analysis of the impact of statistical and semantic features on short texts is necessary to identify the most effective approaches, improve detection accuracy, and ensure that the developed systems can better handle the variety and complexity of informal language. The data used are a public dataset originating from Twitter texts and product review texts in e-commerce. The research utilizes statistical features such as Term Frequency Inverse Document Frequency (TF-IDF) and semantic features such as Bidirectional Encoder Representations from Transformers (BERT). The evaluation results show that using semantic features significantly improves the performance of emotion detection in short texts by 13-24%. It is higher than using statistical features. Deep Learning (DL) algorithms based on neural networks have also been proven to outperform Machine Learning (ML) algorithms in detecting emotions in short text. The experimental results and outlines show the potential directions for future development.