Fake Review Detection Using CatBoost and TF_IDF
Abstract
— The increasing reliance on online reviews to influence purchasing decisions has led to a surge in deceptive practices, including the proliferation of fake reviews. These fraudulent reviews distort customer perception, mislead consumers, and impact businesses unfairly. Traditional rule-based methods and manual detection approaches fail to scale effectively due to the vast volume of user-generated content. In this paper, we present a robust machine learning-based approach for detecting fake reviews using CatBoost, a gradient boosting algorithm optimized for categorical data, in combination with Term Frequency-Inverse Document Frequency (TF-IDF), a widely used text representation technique. Our proposed model preprocesses textual reviews by performing tokenization, stopword removal, lemmatization, and vectorization to convert raw textual data into a meaningful numerical representation. The transformed data is then fed into the CatBoost classifier, which learns patterns and classifies reviews as either genuine or fake. The model is trained on a labeled dataset and evaluated using performance metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that our approach outperforms traditional classifiers such as Naïve Bayes, Logistic Regression, and Support Vector Machines (SVM) in detecting fraudulent reviews with higher accuracy. The integration of TF-IDF for feature extraction ensures that the model effectively captures semantic relevance, while CatBoost’s handling of categorical features and missing values enhances classification performance. Additionally, we developed a Streamlit-based web application to facilitate realtime fake review detection for end users. This research contributes to the growing field of trustworthy AI for ecommerce platforms, enabling businesses and consumers to make informed decisions while mitigating the adverse impact of deceptive reviews. The proposed approach can also be adapted for other domains where textual data integrity is critical, such as news verification and financial fraud detection. Future research can explore the integration of deep learning techniques for enhanced review authenticity assessment.
Keywords
Fake review detection, CatBoost, TF-IDF, Machine Learning, Text Classification, E-commerce Fraud, Sentiment Analysis, Gradient Boosting, Streamlit, NLP