Systematic Literature Review: Machine Learning Methods in Emotion Classification in Textual Data

Emotions are a person's response to an event. Emotions can be expressed verbally or nonverbally. Over time people can express their emotions through social media. Considering that emotion is a reflection of society's response, it is important to classify emotions in society to find out the community's response as information for consideration in decision-making. This study is aimed to identify and analyze the datasets, methods, and evaluation metrics that are being used in the classification of emotional texts in textual data from research data from 2013 to 2022. Based on the inclusion and exclusion design in selecting literature, a total of 50 kinds of literature were used in extracting and synthesizing data. Analysis of the data shows that out of 50 pieces of literature, there are 36 works of literature that use public datasets while 14 kinds of literature use private datasets. In the method of developing models for classifying, the SVM and Naive Bayes models are the most popular among the other models. In evaluating the model, the F-measure or F1-score metric is the most widely used metric compared to other metrics. There are three main contributions identified in this study, namely methods, models, and evaluation