Web mining and sentiment analysis through user comments about KADES application

Various applications and platforms are actively used in our country to combat violence against women. In this context, information technologies and emergency lines stand out as effective communication tools in combating violence by state institutions. In the field of information technologies, the Women's Support Application (KADES), developed in cooperation with the Ministry of Interior and the General Directorate of Security (EGM), stands out as an up-to-date and effective application example that works in integration with security units. In this thesis study, user comments belonging to the KADES application were obtained by data mining method and made suitable for processing in WEKA program with text mining techniques. The data was organized into two different data groups as balanced and unbalanced and the classification process was performed using Naive Bayes, KNN and SMO algorithms. The comments were classified as three-labeled (positive, negative and neutral) and two-labeled (positive and negative), and both groups were evaluated separately as balanced and unbalanced data. According to the analysis results, it was seen that the SMO algorithm was superior to other classifiers with 81.118% classification accuracy in three-labeled unbalanced data. In balanced data with three labels, Naive Bayes algorithm showed good performance with 83.890% accuracy rate, while SMO gave the best result with 84.246% accuracy rate. It was determined that KNN algorithm showed lower performance than other algorithms in the evaluations made using different k values (k=1, k=3 and k=5). In unbalanced data with two labels, SMO was found to be more successful than other methods with 92.359% accuracy rate. Although Naive Bayes achieved successful results with 90.988% accuracy rate in balanced data with two labels, SMO algorithm provided the highest performance with 95.928% accuracy rate. In general, it was observed that classification accuracy increased when balanced data was used.