Large Language Model-Powered Generation of Multilingual Online Consumer Reviews for E-Commerce

The current system in e-commerce platforms accepts product feedback in the form of star ratings and textual reviews. However, due to the short length of these textual reviews, the detailed aspects of products are missed most of the time. To overcome these limitations, the current research proposes a framework based on the Large Language Model (LLaMA 3.2) to generate reviews from ratings of the product's aspects. The proposed framework generates reviews in Hindi-English mixed language to make them more engaging for the Indian population. The results show that AI-generated reviews score better than human-written reviews in clarity (3.55 vs. 3.01), accuracy (3.49 vs. 2.51), and keyword relevance (67.4% vs. 35.7%). These AI-generated reviews were found to be very similar to human-written reviews and were indistinguishable in 59.7% of cases of human evaluation.