Klasifikasi Emosi Multi Label Di Teks Bahasa Inggris Menggunakan Metode Naive Bayes Dan Fitur Term Frequency

Abstract

Emotion classification in text is an important area within the field of Natural Language Processing (NLP). This task becomes increasingly relevant because a single statement or sentence may contain more than one type of emotion simultaneously. This work performs multi-label emotion classification on English text using the Naïve Bayes method with Term Frequency features for five emotion labels, namely anger, fear, joy, sadness, and surprise. Model performance is evaluated using macro precision, recall, and F1-score to assess each emotion label independently. The experiments are conducted with varying values of the Laplace smoothing parameter to obtain the most stable configuration. The experimental results show that the best configuration is achieved with an alpha value of 0,1, and the application of modified stopwords that retain the words not, no, and never yields an average macro F1-score of 0,49. Classification performance varies across emotion labels. The fear emotion achieves the highest F1-score of 0,72, indicating linguistic patterns that are relatively easier to recognize. The sadness emotion attains an F1-score of 0,50, followed by surprise with 0,49 and joy with 0,42. In contrast, the anger emotion records the lowest F1-score of 0,29, indicating a higher level of difficulty in the identification process, primarily due to limited data availability and semantic similarity with other emotions.