Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LEAPS: An LLM-Empowered Adaptive Plugin for Taobao AI Search

Lei Wang, Jinhang Wu, Zhibin Wang, Biye Li, Haiping Hou

TL;DR

LEAPS introduces a non-intrusive, two-plugin framework that bridges complex, high-dimensional conversational user intents with traditional e-commerce search. The upstream Query Expander uses a three-stage, diversity-aware RL regimen to generate complementary rewrites, while the downstream Relevance Verifier fuses multi-source signals and CoT-based reasoning to prune noise and justify selections. Offline experiments show substantial gains in candidate coverage and relevance, complemented by online A/B results that reduce zero-result scenarios and improve CTR, all while preserving short-query retrieval performance. The architecture is deployed at Taobao, demonstrating scalable, low-latency integration with existing backends and significant practical impact for hundreds of millions of users monthly. Future work targets adaptive budget allocation and native multimodal verification to further enhance performance and efficiency.

Abstract

The rapid advancement of large language models has reshaped user search cognition, driving a paradigm shift from discrete keyword-based search to high-dimensional conversational interaction. However, existing e-commerce search architectures face a critical capability deficit in adapting to this change. Users are often caught in a dilemma: precise natural language descriptions frequently trigger zero-result scenarios, while the forced simplification of queries leads to decision overload from noisy, generic results. To tackle this challenge, we propose LEAPS (LLM-Empowered Adaptive Plugin for Taobao AI Search), which seamlessly upgrades traditional search systems via a "Broaden-and-Refine" paradigm. Specifically, it attaches plugins to both ends of the search pipeline: (1) Upstream, a Query Expander acts as an intent translator. It employs a novel three-stage training strategy--inverse data augmentation, posterior-knowledge supervised fine-tuning, and diversity-aware reinforcement learning--to generate adaptive and complementary query combinations that maximize the candidate product set. (2) Downstream, a Relevance Verifier serves as a semantic gatekeeper. By synthesizing multi-source data (e.g., OCR text, reviews) and leveraging chain-of-thought reasoning, it precisely filters noise to resolve selection overload. Extensive offline experiments and online A/B testing demonstrate that LEAPS significantly enhances conversational search experiences. Crucially, its non-invasive architecture preserves established retrieval performance optimized for short-text queries, while simultaneously allowing for low-cost integration into diverse back-ends. Fully deployed on Taobao AI Search since August 2025, LEAPS currently serves hundreds of millions of users monthly.

LEAPS: An LLM-Empowered Adaptive Plugin for Taobao AI Search

TL;DR

LEAPS introduces a non-intrusive, two-plugin framework that bridges complex, high-dimensional conversational user intents with traditional e-commerce search. The upstream Query Expander uses a three-stage, diversity-aware RL regimen to generate complementary rewrites, while the downstream Relevance Verifier fuses multi-source signals and CoT-based reasoning to prune noise and justify selections. Offline experiments show substantial gains in candidate coverage and relevance, complemented by online A/B results that reduce zero-result scenarios and improve CTR, all while preserving short-query retrieval performance. The architecture is deployed at Taobao, demonstrating scalable, low-latency integration with existing backends and significant practical impact for hundreds of millions of users monthly. Future work targets adaptive budget allocation and native multimodal verification to further enhance performance and efficiency.

Abstract

The rapid advancement of large language models has reshaped user search cognition, driving a paradigm shift from discrete keyword-based search to high-dimensional conversational interaction. However, existing e-commerce search architectures face a critical capability deficit in adapting to this change. Users are often caught in a dilemma: precise natural language descriptions frequently trigger zero-result scenarios, while the forced simplification of queries leads to decision overload from noisy, generic results. To tackle this challenge, we propose LEAPS (LLM-Empowered Adaptive Plugin for Taobao AI Search), which seamlessly upgrades traditional search systems via a "Broaden-and-Refine" paradigm. Specifically, it attaches plugins to both ends of the search pipeline: (1) Upstream, a Query Expander acts as an intent translator. It employs a novel three-stage training strategy--inverse data augmentation, posterior-knowledge supervised fine-tuning, and diversity-aware reinforcement learning--to generate adaptive and complementary query combinations that maximize the candidate product set. (2) Downstream, a Relevance Verifier serves as a semantic gatekeeper. By synthesizing multi-source data (e.g., OCR text, reviews) and leveraging chain-of-thought reasoning, it precisely filters noise to resolve selection overload. Extensive offline experiments and online A/B testing demonstrate that LEAPS significantly enhances conversational search experiences. Crucially, its non-invasive architecture preserves established retrieval performance optimized for short-text queries, while simultaneously allowing for low-cost integration into diverse back-ends. Fully deployed on Taobao AI Search since August 2025, LEAPS currently serves hundreds of millions of users monthly.
Paper Structure (35 sections, 5 equations, 4 figures, 2 tables, 1 algorithm)

This paper contains 35 sections, 5 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Query Rewriting
  4. Discriminative Approaches
  5. Generative Approaches
  6. Search Relevance
  7. Traditional Lexical Matching
  8. Pre-trained Language Models (PLMs)
  9. LLMs
  10. Method
  11. Framework Overview
  12. Query Expander
  13. Inverse Data Augmentation
  14. SFT with Posterior Knowledge
  15. RL for Query Diversification
  16. ...and 20 more sections

Figures (4)

  • Figure 1: Schematic overview of the LEAPS framework. The Query Expander employs a three-stage training strategy to generate diverse, system-compliant rewrites that maximize the capture of potential relevant items. The Relevance Verifier functions as a decision-making agent, synthesizing multi-source heterogeneous data and internalized reasoning capabilities to perform precise item filtering. The framework operates as a plug-in, interacting with the black-box search engine solely through query submission and result aggregation.
  • Figure 2: Comparative performance of LEAPS under distinct RL optimization algorithms. Shaded regions denote the standard deviation across 3 independent runs.
  • Figure 3: Impact of the Reasoning Process on LEAPS performance. Shaded regions denote the standard deviation across 3 independent runs.
  • Figure 4: Screenshots from the Taobao AI Search application, illustrating how LEAPS handles complex conversational queries to retrieve compliant products. (a) The user input and the underlying retrieval process. The Expander broadens the scope to form a candidate pool, which is then screened by the Verifier. Validated items (highlighted in green) are integrated into the dual-column feed, while mismatched items (grayed out) are filtered out. (b) The final product list presented to the user after LEAPS processing, demonstrating high alignment with the specific constraints.