Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring Large Language Models for Product Attribute Value Identification

Kassem Sabeh, Mouna Kacimi, Johann Gamper, Robert Litschko, Barbara Plank

TL;DR

The paper tackles Product Attribute Value Identification (PAVI), a challenging task requiring generation of attribute names from product context. It systematically evaluates open-source LLMs (LLaMA-3, Mistral, OLMo) using one-step and two-step prompts, augmented by in-context learning with parametric and non-parametric knowledge and instruction fine-tuning. Key findings show the two-step prompting substantially improves zero-shot performance, while instruction fine-tuning delivers the largest gains when task data is available; dense demonstration retrieval further enhances in-context learning. Practically, these results demonstrate that LLMs can effectively identify product attributes and values with limited task-specific data, enabling improved product search, recommendation, and QA in real-world e-commerce settings. The work also highlights the potential of domain transfer for retrievers and notes limitations related to data splits, over-generation risks, and evaluation scope, pointing to future exploration with proprietary models.

Abstract

Product attribute value identification (PAVI) involves automatically identifying attributes and their values from product information, enabling features like product search, recommendation, and comparison. Existing methods primarily rely on fine-tuning pre-trained language models, such as BART and T5, which require extensive task-specific training data and struggle to generalize to new attributes. This paper explores large language models (LLMs), such as LLaMA and Mistral, as data-efficient and robust alternatives for PAVI. We propose various strategies: comparing one-step and two-step prompt-based approaches in zero-shot settings and utilizing parametric and non-parametric knowledge through in-context learning examples. We also introduce a dense demonstration retriever based on a pre-trained T5 model and perform instruction fine-tuning to explicitly train LLMs on task-specific instructions. Extensive experiments on two product benchmarks show that our two-step approach significantly improves performance in zero-shot settings, and instruction fine-tuning further boosts performance when using training data, demonstrating the practical benefits of using LLMs for PAVI.

Exploring Large Language Models for Product Attribute Value Identification

TL;DR

The paper tackles Product Attribute Value Identification (PAVI), a challenging task requiring generation of attribute names from product context. It systematically evaluates open-source LLMs (LLaMA-3, Mistral, OLMo) using one-step and two-step prompts, augmented by in-context learning with parametric and non-parametric knowledge and instruction fine-tuning. Key findings show the two-step prompting substantially improves zero-shot performance, while instruction fine-tuning delivers the largest gains when task data is available; dense demonstration retrieval further enhances in-context learning. Practically, these results demonstrate that LLMs can effectively identify product attributes and values with limited task-specific data, enabling improved product search, recommendation, and QA in real-world e-commerce settings. The work also highlights the potential of domain transfer for retrievers and notes limitations related to data splits, over-generation risks, and evaluation scope, pointing to future exploration with proprietary models.

Abstract

Product attribute value identification (PAVI) involves automatically identifying attributes and their values from product information, enabling features like product search, recommendation, and comparison. Existing methods primarily rely on fine-tuning pre-trained language models, such as BART and T5, which require extensive task-specific training data and struggle to generalize to new attributes. This paper explores large language models (LLMs), such as LLaMA and Mistral, as data-efficient and robust alternatives for PAVI. We propose various strategies: comparing one-step and two-step prompt-based approaches in zero-shot settings and utilizing parametric and non-parametric knowledge through in-context learning examples. We also introduce a dense demonstration retriever based on a pre-trained T5 model and perform instruction fine-tuning to explicitly train LLMs on task-specific instructions. Extensive experiments on two product benchmarks show that our two-step approach significantly improves performance in zero-shot settings, and instruction fine-tuning further boosts performance when using training data, demonstrating the practical benefits of using LLMs for PAVI.
Paper Structure (33 sections, 7 figures, 9 tables)

This paper contains 33 sections, 7 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Product Attribute Value Extraction.
  4. Product Attribute Value Identification.
  5. LLMs for Feature Extraction.
  6. Methodology
  7. PAVI Task Modelling
  8. One-Step Approach.
  9. Two-Step Approach.
  10. Leveraging LLMs for PAVI
  11. Zero-Shot Settings
  12. In-context Learning
  13. Parametric Knowledge.
  14. Non-Parametric Knowledge.
  15. Instruction Fine-Tuning
  16. ...and 18 more sections

Figures (7)

  • Figure 1: An example of a product listing with tagged attribute-value pairs in the title.
  • Figure 2: Proposed method for using the fine-tuned T5 model as a dense retriever.
  • Figure 3: Zero-shot prompt template for the one-step approach.
  • Figure 4: Zero-shot prompt template for the two-step approach.
  • Figure 5: Template illustrating how self-generated product titles are used in the one-step approach.
  • ...and 2 more figures