Comparative Study of Artificial Intelligence Approaches in Deceptive Opinion Detection

The deceptive opinion spam directly threatens the integrity of online review platforms, which in turn erodes consumer trust and fair competition. In this study, we explore and compare the performance of traditional machine learning models such as Logistic Regression (LR), Support Vector Machine (SVM) as well as recently released advanced deep learning models (BiLSTM, BERT, XLNet, T5, Capsule Networks) to detect deceptive reviews. The research tests different feature representations (unigrams, bigrams, trigrams, TF-IDF) as baseline augmented performance metrics using datasets such as Deceptive Opinion Spam Corpus. The XLNet led all models with the highest performance at 89.69% accuracy and 90.03% F1 score, emerging as the most effective on aggregate. The BERT T5 models provided high precision, BiLSTM served the purpose with balanced performance, and Capsule Networks demonstrated its promise in capturing hierarchical relationship structures in text data. As a result, this study contributes to opinion spam detection by suggesting good solutions and open tracks for future refinements, such as hybrid model development or domain-specific fine-tuning.