Towards Better Search with Domain-Aware Text Embeddings for C2C Marketplaces

TL;DR

This paper presents a domain-aware Japanese text-embedding approach for C2C marketplaces, improving search relevance and efficiency through specialized fine-tuning and compact embedding techniques, validated by offline and online experiments.

Contribution

Introduces a novel domain-aware Japanese text-embedding method with role-specific modeling and Matryoshka truncation for improved C2C marketplace search.

Findings

Significant offline performance gains over generic encoders.

Enhanced handling of proper nouns and marketplace-specific terms.

Online A/B tests show increased revenue and search efficiency.

Abstract

Consumer-to-consumer (C2C) marketplaces pose distinct retrieval challenges: short, ambiguous queries; noisy, user-generated listings; and strict production constraints. This paper reports our experiment to build a domain-aware Japanese text-embedding approach to improve the quality of search at Mercari, Japan's largest C2C marketplace. We experimented with fine-tuning on purchase-driven query-title pairs, using role-specific prefixes to model query-item asymmetry. To meet production constraints, we apply Matryoshka Representation Learning to obtain compact, truncation-robust embeddings. Offline evaluation on historical search logs shows consistent gains over a strong generic encoder, with particularly large improvements when replacing PCA compression with Matryoshka truncation. A manual assessment further highlights better handling of proper nouns, marketplace-specific semantics, and term-importance alignment. Additionally, an initial online A/B test demonstrates statistically significant improvements in revenue per user and search-flow efficiency, with transaction frequency maintained. Results show that domain-aware embeddings improve relevance and efficiency at scale and form a practical foundation for richer LLM-era search experiences.