Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ChoirRec: Semantic User Grouping via LLMs for Conversion Rate Prediction of Low-Activity Users

Dakai Zhai, Jiong Gao, Boya Du, Junwei Xu, Qijie Shen, Jialin Zhu, Yuning Jiang

TL;DR

ChoirRec addresses CVR prediction for low-activity users by constructing high-quality semantic user groups via LLMs and leveraging them through a group-aware, dual-channel architecture. The method combines semantic-profile synthesis, hierarchical grouping with RQ-KMeans, and multi-faceted group priors (ID fusion, attribute completion, and group sequences) into a dual-channel predictor with asymmetric information injection and gated knowledge distillation. Empirical results on Taobao demonstrate offline GAUC gains for low-activity users and substantial online improvements in orders and GMV, validating the practical value of cross-user semantic transfer for sparse signals. The findings suggest that semantically grounded group knowledge can effectively bridge long-tail user gaps and improve real-world recommender performance.

Abstract

Accurately predicting conversion rates (CVR) for low-activity users remains a fundamental challenge in large-scale e-commerce recommender systems. Existing approaches face three critical limitations: (i) reliance on noisy and unreliable behavioral signals; (ii) insufficient user-level information due to the lack of diverse interaction data; and (iii) a systemic training bias toward high-activity users that overshadows the needs of low-activity users. To address these challenges, we propose ChoirRec, a novel framework that leverages the semantic capabilities of Large Language Models (LLMs) to construct semantic user groups and enhance CVR prediction for low-activity users. With a dual-channel architecture designed for robust cross-user knowledge transfer, ChoirRec comprises three components: (i) a Semantic Group Generation module that utilizes LLMs to form reliable, cross-activity user clusters, thereby filtering out noisy signals; (ii) a Group-aware Hierarchical Representation module that enriches sparse user embeddings with informative group-level priors to mitigate data insufficiency; and (iii) a Group-aware Multi-granularity Modual that employs a dual-channel architecture and adaptive fusion mechanism to ensure effective learning and utilization of group knowledge. We conduct extensive offline and online experiments on Taobao, a leading industrial-scale e-commerce platform. ChoirRec improves GAUC by 1.16\% in offline evaluations, while online A/B testing reveals a 7.24\% increase in order volume, highlighting its substantial practical value in real-world applications.

ChoirRec: Semantic User Grouping via LLMs for Conversion Rate Prediction of Low-Activity Users

TL;DR

ChoirRec addresses CVR prediction for low-activity users by constructing high-quality semantic user groups via LLMs and leveraging them through a group-aware, dual-channel architecture. The method combines semantic-profile synthesis, hierarchical grouping with RQ-KMeans, and multi-faceted group priors (ID fusion, attribute completion, and group sequences) into a dual-channel predictor with asymmetric information injection and gated knowledge distillation. Empirical results on Taobao demonstrate offline GAUC gains for low-activity users and substantial online improvements in orders and GMV, validating the practical value of cross-user semantic transfer for sparse signals. The findings suggest that semantically grounded group knowledge can effectively bridge long-tail user gaps and improve real-world recommender performance.

Abstract

Accurately predicting conversion rates (CVR) for low-activity users remains a fundamental challenge in large-scale e-commerce recommender systems. Existing approaches face three critical limitations: (i) reliance on noisy and unreliable behavioral signals; (ii) insufficient user-level information due to the lack of diverse interaction data; and (iii) a systemic training bias toward high-activity users that overshadows the needs of low-activity users. To address these challenges, we propose ChoirRec, a novel framework that leverages the semantic capabilities of Large Language Models (LLMs) to construct semantic user groups and enhance CVR prediction for low-activity users. With a dual-channel architecture designed for robust cross-user knowledge transfer, ChoirRec comprises three components: (i) a Semantic Group Generation module that utilizes LLMs to form reliable, cross-activity user clusters, thereby filtering out noisy signals; (ii) a Group-aware Hierarchical Representation module that enriches sparse user embeddings with informative group-level priors to mitigate data insufficiency; and (iii) a Group-aware Multi-granularity Modual that employs a dual-channel architecture and adaptive fusion mechanism to ensure effective learning and utilization of group knowledge. We conduct extensive offline and online experiments on Taobao, a leading industrial-scale e-commerce platform. ChoirRec improves GAUC by 1.16\% in offline evaluations, while online A/B testing reveals a 7.24\% increase in order volume, highlighting its substantial practical value in real-world applications.

Paper Structure

This paper contains 28 sections, 8 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Knowledge Transfer via Data and Representation Enhancement
  4. Specialized Model Architectures for Data Sparsity
  5. LLM-Enhanced Recommendation
  6. Problem Formulation
  7. Methodology
  8. Overview
  9. Hierarchical Semantic User Group Generation
  10. Semantic Profile Synthesis via LLM
  11. Hierarchical Group Construction
  12. Group-aware Hierarchical Representation
  13. Hierarchical Group ID Fusion
  14. Group Attribute Completion
  15. Group Behavioral Sequence Construction
  16. ...and 13 more sections

Figures (6)

  • Figure 1: User activity and embedding distributions
  • Figure 2: The overall architecture of the ChoirRec framework.
  • Figure 3: The construction of the Group-aware Hierarchical Representation system.
  • Figure 4: The architecture of the Group-aware Multi-granularity Module
  • Figure 5: Relative GAUC improvement of ChoirRec over the Base Model across five user activity levels, ordered by historical purchase frequency (Level 1 = lowest activity, Level 5 = highest).
  • ...and 1 more figures