Improvement of feature engineering on emotion detection from textual data

This paper introduces a new method for selecting terms in the field of emotion recognition from text. Instead of focusing solely on very common or very rare terms, this approach considers moderately frequent terms as well. The idea is that these moderately frequent terms might also contain important information for distinguishing between emotions. Compared to traditional methods like Chi-Square and Gini-Text, this new approach performs better in many cases. To represent documents, the bag-of-words approach is used, where each document is represented by a vector. In this vector, each selected term is given a weight of 1 if it appears in the document, and 0 if it does not. Importantly, this new method includes terms that are not selected by Chi-Square and Gini-Text. Experiments conducted on a standard dataset demonstrate that including moderately frequent terms improves the accuracy of emotion recognition. This improvement is evident in terms of accuracy scores.