Author(s): Ryan Daniel Daneshyari

Abstract: This research paper delves into the application of fine-tuning techniques on the Llama model, a cutting-edge open source Large Language Model (LLM) that is used to predict departments within the online grocery shopping context. Department prediction, a critical task in e-commerce, aims to classify products into categories such as "Produce," "Dairy," or "Snacks," making it easier for customers to browse and receive accurate product recommendations. Leveraging an extensive dataset sourced from Instacart, a popular online grocery delivery service, this study utilizes product names and descriptions to refine the Llama model's classification capabilities. This dataset is well-suited for department prediction, containing valuable linguistic nuances specific to grocery items that are challenging for general NLP models to interpret. The proposed methodology includes pre-training the Llama model on a diverse corpus of textual data to establish a broad linguistic foundation, followed by fine-tuning on the Instacart dataset to adapt the model to the unique vocabulary and contextual patterns of grocery items. Fine-tuning, a crucial step in model customization, allows us to tailor the Llama model to recognize product attributes and assign them accurately to their respective departments. This approach not only enhances classification accuracy but also sheds light on the adaptability of NLP models to specialized domains like online grocery shopping. To evaluate the effectiveness of the fine-tuned Llama model, a rigorous experimentation has been conducted and its performance is compared against baseline models and conventional classification methods. The promising findings indicate that fine-tuning significantly boosts the model’s ability to accurately predict departments, outperforming traditional approaches in categorizing complex product names and descriptions. The implications of this research extend beyond department prediction and provide insights into the broader applicability of advanced LLM models for e-commerce. By improving product categorization, the proposed study contributes to optimizing user experience, refining recommendation systems, and enhancing operational efficiency within online retail platforms, ultimately supporting better customer engagement and satisfaction.

DOI: 10.33545/26633582.2024.v6.i2c.141

Pages: 199-203 | Views: 1075 | Downloads: 569

Download Full Article: Click Here