Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Query Brand Entity Linking in E-Commerce Search

Dong Liu, Sreyashi Nag

TL;DR

This work tackles brand entity linking in e-commerce search by presenting a two-stage approach (NER-based detection plus lexical or semantic matching with product-type filtering) and an end-to-end extreme multi-class framework (Q2E-PECOS) that directly predicts brand entities from queries. It introduces MetaTS-NER, Q2PT, and PECOS as core preliminaries, and demonstrates that fusing end-to-end and two-stage predictions yields higher coverage and recall while maintaining precision. Extensive offline benchmarks and online A/B tests show the PECOS-based fusion achieves favorable trade-offs, improving brand recall and key engagement metrics in real-world settings. The study highlights practical deployment considerations, including NIL handling for non-branded queries and multilingual data, with concrete gains in both user experience and business outcomes.

Abstract

In this work, we address the brand entity linking problem for e-commerce search queries. The entity linking task is done by either i)a two-stage process consisting of entity mention detection followed by entity disambiguation or ii) an end-to-end linking approaches that directly fetch the target entity given the input text. The task presents unique challenges: queries are extremely short (averaging 2.4 words), lack natural language structure, and must handle a massive space of unique brands. We present a two-stage approach combining named-entity recognition with matching, and a novel end-to-end solution using extreme multi-class classification. We validate our solutions by both offline benchmarks and the impact of online A/B test.

Query Brand Entity Linking in E-Commerce Search

TL;DR

This work tackles brand entity linking in e-commerce search by presenting a two-stage approach (NER-based detection plus lexical or semantic matching with product-type filtering) and an end-to-end extreme multi-class framework (Q2E-PECOS) that directly predicts brand entities from queries. It introduces MetaTS-NER, Q2PT, and PECOS as core preliminaries, and demonstrates that fusing end-to-end and two-stage predictions yields higher coverage and recall while maintaining precision. Extensive offline benchmarks and online A/B tests show the PECOS-based fusion achieves favorable trade-offs, improving brand recall and key engagement metrics in real-world settings. The study highlights practical deployment considerations, including NIL handling for non-branded queries and multilingual data, with concrete gains in both user experience and business outcomes.

Abstract

In this work, we address the brand entity linking problem for e-commerce search queries. The entity linking task is done by either i)a two-stage process consisting of entity mention detection followed by entity disambiguation or ii) an end-to-end linking approaches that directly fetch the target entity given the input text. The task presents unique challenges: queries are extremely short (averaging 2.4 words), lack natural language structure, and must handle a massive space of unique brands. We present a two-stage approach combining named-entity recognition with matching, and a novel end-to-end solution using extreme multi-class classification. We validate our solutions by both offline benchmarks and the impact of online A/B test.

Paper Structure

This paper contains 22 sections, 5 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. MetaTS-NER
  5. Q2PT
  6. PECOS
  7. Brand Entity Linking via Two-Stage Framework
  8. MetaTS-NER $+$ Exact Lexical Matching
  9. NER $+$ PECOS Semantic Matching
  10. Brand Entity Linking via Extreme Multi-Class Classification Framework
  11. Fusion of the Q2E-PECOS and Exact Lexical Match Prediction
  12. Experiments
  13. Datasets
  14. Training Dataset
  15. Test Dataset
  16. ...and 7 more sections

Figures (3)

  • Figure 1: A NER based two-stage framework for brand entity linking
  • Figure 2: A NER-and-PECOS-based two-stage framework for brand entity linking. The framework consists of: (1) Brand mention detection using MetaTS-NER, (2) Semantic matching using PECOS to map mentions to brand entities, and (3) Product type-based filtering to disambiguate between multiple candidate entities.
  • Figure 3: (a) A PECOS based End-to-End framework for brand entity linking. The framework maps queries to brand entities using extreme multi-class classification, followed by product type-based filtering. (b) Fusion of one-stage and two-stage approaches, combining their predictions with priority given to lexical matching results.