Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLaSA: Large Language and E-Commerce Shopping Assistant

Shuo Zhang, Boci Peng, Xinping Zhao, Boren Hu, Yun Zhu, Yanjia Zeng, Xuming Hu

TL;DR

This work tackles the challenge of building scalable e commerce shopping assistants by leveraging large language models with domain specific instruction tuning. It introduces EshopInstruct a 65k sample instruction dataset and LLaSA an instruction tuned shopping assistant. The study uses LoRA fine tuning and GPTQ quantization to enable efficient inference while maintaining capabilities across multilingual and domain tasks. On the Amazon KDD Cup 2024 ShopBench benchmark LLaSA achieves 3rd place overall and top-5 in all tracks, validating LLMs as effective e commerce assistants with practical inference strategies.

Abstract

The e-commerce platform has evolved rapidly due to its widespread popularity and convenience. Developing an e-commerce shopping assistant for customers is crucial to aiding them in quickly finding desired products and recommending precisely what they need. However, most previous shopping assistants face two main problems: (1) task-specificity, which necessitates the development of different models for various tasks, thereby increasing development costs and limiting effectiveness; and (2) poor generalization, where the trained model performs inadequately on up-to-date products. To resolve these issues, we employ Large Language Models (LLMs) to construct an omnipotent assistant, leveraging their adeptness at handling multiple tasks and their superior generalization capability. Nonetheless, LLMs lack inherent knowledge of e-commerce concepts. To address this, we create an instruction dataset comprising 65,000 samples and diverse tasks, termed as EshopInstruct. Through instruction tuning on our dataset, the assistant, named LLaSA, demonstrates the potential to function as an omnipotent assistant. Additionally, we propose various inference optimization strategies to enhance performance with limited inference resources. In the Amazon KDD Cup 2024 Challenge, our proposed method, LLaSA, achieved an overall ranking of 3rd place on ShopBench, including 57 tasks and approximately 20,000 questions, and we secured top-5 rankings in each track, especially in track4, where we achieved the best performance result among all student teams. Our extensive practices fully demonstrate that LLMs possess the great potential to be competent e-commerce shopping assistants.

LLaSA: Large Language and E-Commerce Shopping Assistant

TL;DR

This work tackles the challenge of building scalable e commerce shopping assistants by leveraging large language models with domain specific instruction tuning. It introduces EshopInstruct a 65k sample instruction dataset and LLaSA an instruction tuned shopping assistant. The study uses LoRA fine tuning and GPTQ quantization to enable efficient inference while maintaining capabilities across multilingual and domain tasks. On the Amazon KDD Cup 2024 ShopBench benchmark LLaSA achieves 3rd place overall and top-5 in all tracks, validating LLMs as effective e commerce assistants with practical inference strategies.

Abstract

The e-commerce platform has evolved rapidly due to its widespread popularity and convenience. Developing an e-commerce shopping assistant for customers is crucial to aiding them in quickly finding desired products and recommending precisely what they need. However, most previous shopping assistants face two main problems: (1) task-specificity, which necessitates the development of different models for various tasks, thereby increasing development costs and limiting effectiveness; and (2) poor generalization, where the trained model performs inadequately on up-to-date products. To resolve these issues, we employ Large Language Models (LLMs) to construct an omnipotent assistant, leveraging their adeptness at handling multiple tasks and their superior generalization capability. Nonetheless, LLMs lack inherent knowledge of e-commerce concepts. To address this, we create an instruction dataset comprising 65,000 samples and diverse tasks, termed as EshopInstruct. Through instruction tuning on our dataset, the assistant, named LLaSA, demonstrates the potential to function as an omnipotent assistant. Additionally, we propose various inference optimization strategies to enhance performance with limited inference resources. In the Amazon KDD Cup 2024 Challenge, our proposed method, LLaSA, achieved an overall ranking of 3rd place on ShopBench, including 57 tasks and approximately 20,000 questions, and we secured top-5 rankings in each track, especially in track4, where we achieved the best performance result among all student teams. Our extensive practices fully demonstrate that LLMs possess the great potential to be competent e-commerce shopping assistants.
Paper Structure (18 sections, 3 figures, 4 tables)

This paper contains 18 sections, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Background
  3. Datasets Description
  4. Task Description
  5. TRAINING DATASET CONSTRUCTION
  6. Development Set Analysis
  7. Shopping Concept Understanding
  8. Shopping Knowledge Reasoning
  9. User Behavior Alignment
  10. Multi-Lingual Abilities
  11. External Datasets
  12. Training Data Construction Strategy
  13. INSTRUCTION TUNING
  14. INFERENCE
  15. Quantification
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Construction pipeline of EshopInstruct. We design three strategies for building the EshopInstruct dataset: generating data from seed data, extracting data from publicly available ECInstruct, and designing new tasks to generate data. Based on these strategies, we obtained 65k data points.
  • Figure 2: The data distribution of development set, including four important shopping skills: (1) Shopping Concept Understanding; (2) Shopping Knowledge Reasoning; (3) User Behavior Alignment; (4) Multi-Lingual Abilities, and five abilities: (i) Generation; (ii) Ranking; (iii) Retrieval; (iv) Multiple-Choice; (v) NER.
  • Figure 3: The overall inference framework of our solution. In our prompt construction, we enhanced the model's reasoning capabilities by incorporating few-shot examples retrieved using queries, the "read again" technique, and chain-of-thought reasoning. Additionally, for Qwen2-72B, we applied GPTQ quantization, enabling it to run efficiently on limited resources.