Leveraging Large Language Models and RAPIDS AI to Detect Fraudulent AI-Generated Marketplace Reviews

Consumer confidence is, in the present time, a dilemma given the steadily rising number of deceptive and inaccurate AI-generated reviews on internet marketplaces. There is an urgent need for a thorough dataset, which is nowhere to be found, to address the pervasiveness of fraudulent reviews. In order to close this gap, in this work, we propose relevant detection techniques using GPU-accelerated models from the RAPIDS AI library and new age Large Language Models like GPT-2 and BERT. By carefully constructing a unique dataset of about 54,000 reviews, we have combined human-authored reviews gleaned from web scraping with AI-generated reviews produced with the Chat-GPT 3.5 model, which supports the representative spectrum across all review polarities and offers a solid basis for precise analysis. We show significant accuracy in differentiating between reviews written by humans and those produced by AI and provide an all-inclusive solution to this new issue by using sophisticated LLMs and RAPIDS AI models. Moreover, we also compare the approaches to find the most successful one. This approach not only tackles the problem of fraudulent reviews but also establishes a new benchmark for keeping online marketplaces morally sound.