Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ask Me What You Need: Product Retrieval using Knowledge from GPT-3

Su Young Kim, Hyeonjin Park, Kyuyong Shin, Kyung-Min Kim

TL;DR

This work introduces a GPT-3–driven product retrieval framework that treats GPT-3 as an implicit knowledge base and uses p-tuning to prompt relevant product-category knowledge from natural-language queries. A two-stage pipeline first retrieves top-K product categories via GPT-3 prompts and token-score aggregation, then ranks candidate products with a BERT-based encoder trained through BCE loss. Across two real-world datasets and a public benchmark, the approach, especially with a 13B GPT-3 model, outperforms traditional lexical and embedding baselines and shows strong cold-start performance. The study highlights the value of prompt-based knowledge integration for intent-driven retrieval and suggests future personalization by incorporating user behavior signals.

Abstract

As online merchandise become more common, many studies focus on embedding-based methods where queries and products are represented in the semantic space. These methods alleviate the problem of vocab mismatch between the language of queries and products. However, past studies usually dealt with queries that precisely describe the product, and there still exists the need to answer imprecise queries that may require common sense knowledge, i.e., 'what should I get my mom for Mother's Day.' In this paper, we propose a GPT-3 based product retrieval system that leverages the knowledge-base (KB) of GPT-3 for question answering; users do not need to know the specific illustrative keywords for a product when querying. Our method tunes prompt tokens of GPT-3 to prompt knowledge and render answers that are mapped directly to products without further processing. Our method shows consistent performance improvement on two real-world and one public dataset, compared to the baseline methods. We provide an in-depth discussion on leveraging GPT-3 knowledge into a question answering based retrieval system.

Ask Me What You Need: Product Retrieval using Knowledge from GPT-3

TL;DR

This work introduces a GPT-3–driven product retrieval framework that treats GPT-3 as an implicit knowledge base and uses p-tuning to prompt relevant product-category knowledge from natural-language queries. A two-stage pipeline first retrieves top-K product categories via GPT-3 prompts and token-score aggregation, then ranks candidate products with a BERT-based encoder trained through BCE loss. Across two real-world datasets and a public benchmark, the approach, especially with a 13B GPT-3 model, outperforms traditional lexical and embedding baselines and shows strong cold-start performance. The study highlights the value of prompt-based knowledge integration for intent-driven retrieval and suggests future personalization by incorporating user behavior signals.

Abstract

As online merchandise become more common, many studies focus on embedding-based methods where queries and products are represented in the semantic space. These methods alleviate the problem of vocab mismatch between the language of queries and products. However, past studies usually dealt with queries that precisely describe the product, and there still exists the need to answer imprecise queries that may require common sense knowledge, i.e., 'what should I get my mom for Mother's Day.' In this paper, we propose a GPT-3 based product retrieval system that leverages the knowledge-base (KB) of GPT-3 for question answering; users do not need to know the specific illustrative keywords for a product when querying. Our method tunes prompt tokens of GPT-3 to prompt knowledge and render answers that are mapped directly to products without further processing. Our method shows consistent performance improvement on two real-world and one public dataset, compared to the baseline methods. We provide an in-depth discussion on leveraging GPT-3 knowledge into a question answering based retrieval system.
Paper Structure (16 sections, 5 equations, 1 figure)

This paper contains 16 sections, 5 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Methodology
  3. Retrieval model
  4. Ranking Model
  5. Experiments
  6. Datasets and Evaluation Metrics
  7. Methods Compared
  8. Results
  9. Retrieval Performance
  10. Product Retrieval
  11. Public Dataset
  12. Effective Tuning Method for Knowledge Retrieval
  13. Effective Tuning Method for Knowledge Retrieval
  14. Influence of Language Model Size
  15. Performance on Cold-Start Problem
  16. ...and 1 more sections

Figures (1)

  • Figure 1: The overview of the proposed product retrieval system. (a) The retrieval model takes a query and selects top-$K$ categories according to the category score. (b) The top-$K$ categories are mapped to candidate products using category-to-product mapping table. (c) Lastly, the candidate products are ranked according to the ranking model.