Artifical Intelligence-Based Sentiment Analysis in Project Tracking Software Jira
Merve Emen, Mehmet Zeki Konyar · 2024
This study, sentiment classification was performed using data obtained from Jira comments. Classification was conducted using the text part of the dataset with 5869 rows obtained from Kaggle. To convert verbal expressions into numerical representations, Count Vectorizer and TFIDF methods were used. The algorithms employed include Support Vector Machines, Decision Tree, Random Forest, XGBoost, K-Nearest Neighbors, and Naive Bayes. Hyperparameter optimization was performed using Gridsearch for the K-Nearest Neighbors, XGBoost, Random Forest, and Support Vector Machines algorithms. According to experimental results, the highest classification accuracy was achieved with the Count Vectorizer method using the XGBoost algorithm, at 86.4%. The average accuracy of algorithms using the Count Vectorizer method was found to be 76.2%, while the average accuracy of algorithms using the TFIDF method was 76.9%, and the average accuracy of algorithms with Gridsearch optimization was 77.7%.