COMPUTER-GENERATED REVIEW CLASSIFICATION USING SUPPORT VECTOR MACHINE WITH FEATURE SELECTION

• Main Author: Caronica Jonur, Rachita
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85290
• Institution: Institut Teknologi Bandung

The growth of e-commerce has led many consumers to shift towards online shopping. In an effort to reduce the risk of making purchasing mistakes, consumers often conduct research by reading reviews of the products they intend to buy. However, not all available reviews can be relied upon as credible sources of information. With advancements in technology, computer- generated reviews have begun to resemble human-written reviews, creating challenges in distinguishing between the two. This study aims to identify features that differentiate human- generated reviews from computer-generated reviews and to develop a classification model using the Support Vector Machine (SVM) algorithm. The developed model is then compared with the classification results produced by GPT-4. In developing the model, tests were conducted on five main features with a total of 34 sub-features, along with hyperparameter tuning to determine the most effective kernel type for influencing SVM performance. The results of the study show that the SVM model with a Radial Basis Function (RBF) kernel, combined with sentiment, syntactic, repetitiveness, and similarity features, provides the best performance with an accuracy of 85.98%, precision of 86.44%, recall of 85.55%, F1 score of 85.99%, and AUC.