Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Alibaba International E-commerce Product Search Competition DILAB Team Technical Report

Hyewon Lee, Junghyun Oh, Minkyung Song, Soyoung Park, Seunghoon Han

TL;DR

The paper tackles multilingual e-commerce product search by formulating two relevance tasks—Query-Category (QC) and Query-Item (QI)—across many languages. It introduces a four-stage pipeline combining Quality Refinement, Tagging, In-Context Learning with Chain-of-Thought, and Unified LLM Training to improve cross-lingual relevance, achieving a top 5 finish with an overall score of 0.8819. Key contributions include data-quality refinement via self-evaluation, task-specific tagging (LT, HCT, SIT, DG), and the integration of ICL/CoT with LoRA-finetuned backbones across QC and QI. The results demonstrate robust cross-language performance and practical applicability for global e-commerce search, supported by ablation analyses and a public codebase for reproducibility.

Abstract

This study presents the multilingual e-commerce search system developed by the DILAB team, which achieved 5th place on the final leaderboard with a competitive overall score of 0.8819, demonstrating stable and high-performing results across evaluation metrics. To address challenges in multilingual query-item understanding, we designed a multi-stage pipeline integrating data refinement, lightweight preprocessing, and adaptive modeling. The data refinement stage enhanced dataset consistency and category coverage, while language tagging and noise filtering improved input quality. In the modeling phase, multiple architectures and fine-tuning strategies were explored, and hyperparameters optimized using curated validation sets to balance performance across query-category (QC) and query-item (QI) tasks. The proposed framework exhibited robustness and adaptability across languages and domains, highlighting the effectiveness of systematic data curation and iterative evaluation for multilingual search systems. The source code is available at https://github.com/2noweyh/DILAB-Alibaba-Ecommerce-Search.

Alibaba International E-commerce Product Search Competition DILAB Team Technical Report

TL;DR

The paper tackles multilingual e-commerce product search by formulating two relevance tasks—Query-Category (QC) and Query-Item (QI)—across many languages. It introduces a four-stage pipeline combining Quality Refinement, Tagging, In-Context Learning with Chain-of-Thought, and Unified LLM Training to improve cross-lingual relevance, achieving a top 5 finish with an overall score of 0.8819. Key contributions include data-quality refinement via self-evaluation, task-specific tagging (LT, HCT, SIT, DG), and the integration of ICL/CoT with LoRA-finetuned backbones across QC and QI. The results demonstrate robust cross-language performance and practical applicability for global e-commerce search, supported by ablation analyses and a public codebase for reproducibility.

Abstract

This study presents the multilingual e-commerce search system developed by the DILAB team, which achieved 5th place on the final leaderboard with a competitive overall score of 0.8819, demonstrating stable and high-performing results across evaluation metrics. To address challenges in multilingual query-item understanding, we designed a multi-stage pipeline integrating data refinement, lightweight preprocessing, and adaptive modeling. The data refinement stage enhanced dataset consistency and category coverage, while language tagging and noise filtering improved input quality. In the modeling phase, multiple architectures and fine-tuning strategies were explored, and hyperparameters optimized using curated validation sets to balance performance across query-category (QC) and query-item (QI) tasks. The proposed framework exhibited robustness and adaptability across languages and domains, highlighting the effectiveness of systematic data curation and iterative evaluation for multilingual search systems. The source code is available at https://github.com/2noweyh/DILAB-Alibaba-Ecommerce-Search.
Paper Structure (19 sections, 6 figures, 6 tables)

This paper contains 19 sections, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Quality Refinement (QR)
  4. LLM Fine-Tuning and Prediction
  5. Self-Evaluation and Quality Cleaning
  6. Tagging
  7. In-Context Learning and Chain-of-Thought
  8. LLM Training and Prediction
  9. Experiments
  10. Requirements
  11. Implementation
  12. Code Structure
  13. Data Preprocessing
  14. Training
  15. Prediction
  16. ...and 4 more sections

Figures (6)

  • Figure 1: Label ratio distribution across languages in the QC and QI training datasets.
  • Figure 2: Overview of the proposed DILAB framework for the QC and QI tasks.
  • Figure 3: Representative QC prompt structure.
  • Figure 4: Representative QI prompt structure.
  • Figure 5: Project directory tree with descriptions.
  • ...and 1 more figures