Automated Product Description Generator Using GPT-Neo: Leveraging Transformer-Based Language Models on Amazon Review Dataset
Bryan Felix, Alexander Agung Santoso Gunawan, Derwin Suhartono · 2024
In this modern era, technological advancements are progressing rapidly, leading to many activities being conducted online. One of the sectors most significantly impacted is the trade sector, where one contributing factor is the emergence of e-commerce. This presents a unique challenge, especially for sellers who wish to conduct business on such an e-commerce platform, where the time required to upload a product is not insignificant. Therefore, this research aims to develop a transformer-based model using the Amazon review dataset. This study compares baseline GPT-2 model with the proposed GPT-Neo model. The results indicate that the proposed GPT-Neo model performs better. The GPT-Neo model achieved the best scores on the test data using a temperature parameter of 0.7 and a top_k of 50, with an average BLEU score of 43.10% and a ROUGE-L score of 11.93%, showcasing its capacity to create more coherent and contextually correct product descriptions. Furthermore, it improves businesses' overall e-commerce experience.