[論文レビュー] Leveraging Large Language Models for Enhanced Product Descriptions in eCommerce

Zhou Jiang-hong, Bo Liu+3 more|arXiv (Cornell University)|Oct 24, 2023

Advanced Text Analysis Techniques被引用数 8

ひとこと要約

The paper fine-tunes LLAMA 2.0 7B on Walmart product descriptions to automate and enhance eCommerce descriptions, evaluating with NDCG, CTR, and human judgments, showing scalability and business impact.

ABSTRACT

In the dynamic field of eCommerce, the quality and comprehensiveness of product descriptions are pivotal for enhancing search visibility and customer engagement. Effective product descriptions can address the 'cold start' problem, align with market trends, and ultimately lead to increased click-through rates. Traditional methods for crafting these descriptions often involve significant human effort and may lack both consistency and scalability. This paper introduces a novel methodology for automating product description generation using the LLAMA 2.0 7B language model. We train the model on a dataset of authentic product descriptions from Walmart, one of the largest eCommerce platforms. The model is then fine-tuned for domain-specific language features and eCommerce nuances to enhance its utility in sales and user engagement. We employ multiple evaluation metrics, including NDCG, customer click-through rates, and human assessments, to validate the effectiveness of our approach. Our findings reveal that the system is not only scalable but also significantly reduces the human workload involved in creating product descriptions. This study underscores the considerable potential of large language models like LLAMA 2.0 7B in automating and optimizing various facets of eCommerce platforms, offering significant business impact, including improved search functionality and increased sales.

研究の動機と目的

• Motivate improved search visibility and customer engagement through better product descriptions.
• Address the cold-start problem by leveraging engagement data to generate descriptions.
• Develop an automated, scalable description-generation pipeline for real-world eCommerce.
• Fine-tune a large language model to produce descriptions optimized for engagement metrics.

提案手法

• Segment product descriptions into five aspects: language appeal, factual information, product dimensions, unique attributes, and brand guarantees.
• Fine-tune LLAMA 2.0 7B on each aspect using a CTR-guided objective that combines language-model likelihood with CTR loss.
• Use a composite objective: L(theta)=lambda*L_NLL(theta)+(1-lambda)*L_CTR(theta).
• Model CTR with logistic regression on description features: y_d = sigma(w^T x_d + b) and L_CTR as negative log-likelihood of observed clicks.
• Evaluate by assembling aspect content into full descriptions and measuring with NDCG, CTR, and human assessments.

実験結果

リサーチクエスチョン

• RQ1 How effective is LLAMA 2.0 7B at generating engaging, high-quality product descriptions for eCommerce?
• RQ2 What is the impact of aspect-based fine-tuning on engagement metrics like CTR and search relevance?
• RQ3 How do automated descriptions compare to original descriptions in BM25 relevance and human-perceived quality?

主な発見

Aspect	Original Description	Enhanced Description
Appealing Introduction	Get in on a cool-casual style with Terra & Sky’s Jeggings for Women.	Dive into an effortlessly chic style with Terra & Sky’s exclusive Women’s Jeggings, tailored just for you.
Factual Information	Material: 61% Cotton/24% Polyester/14% Rayon/1% Spandex. Care: Machine washable. Country of Origin: Imported.	Crafted with a premium blend of 61% Cotton, 24% Polyester, 14% Rayon, and 1% Spandex, these jeggings assure durability and longevity.
Product Dimensions	Size: Model is 5’11” and is wearing a size 1X. Fit: Skinny fit. Rise and Inseam: High rise; 28” inseam.	Specifically designed for a flattering silhouette, these jeggings come in a high-rise style with a 28” inseam.
Unique Attributes	The inner elasticized waist and stretch denim fabric provide a comfortable fit.	Stand out with the jeggings’ inner elasticized waist and stretch denim fabric.
Brand-Related Guarantees	Only at Walmart.	Terra & Sky redefines elegance, exclusive at Walmart.
Pairing Tip	Pair these with your favorite graphic tee.	Team up these jeggings with a chic top.
Series	Women’s Plus Size Jeans from Terra & Sky	Part of the Women’s Plus Size Jeans collection by Terra & Sky.

• BM25 score improved from 66.44 (bottom 50%) to 78.65 (enhanced).
• Human-evaluated NDCG improved from 0.68 to 0.76.
• Top-50% benchmark remained at 82.76 for BM25, indicating gap but narrowing.
• Ablation shows optimal lambda around 0.429 for BM25 performance.
• Descriptions enhanced across five aspects, with clearer, more appealing language and maintained factual content.