Transfer Learning for E-commerce Query Product Type Prediction

TL;DR

The paper addresses Q2PT in multi-locale e-commerce, where short, ambiguous queries and thousands of product types hinder predictive accuracy, especially in low-resource locales. It proposes a unified multilingual Q2PT framework with three variants (NU, U_ag, U_aw) leveraging transfer learning and locale conditioning, evaluated on 20 locales and 1414 product types using a DistilBERT encoder. Results show that unified models outperform per-locale baselines, with locale-aware conditioning (U_aw) delivering the strongest gains, particularly for low-resource markets and long-tail product types, supported by both human and large-scale automatic evaluations and an online user study. The work provides actionable guidance for global query understanding systems, demonstrating that cross-locale knowledge transfer can reduce data demands and infrastructure while improving relevance and user engagement. It also offers a detailed analysis of locale differences and categorizes sources of divergence to inform future cross-cultural NLP applications.

Abstract

Getting a good understanding of the customer intent is essential in e-commerce search engines. In particular, associating the correct product type to a search query plays a vital role in surfacing correct products to the customers. Query product type classification (Q2PT) is a particularly challenging task because search queries are short and ambiguous, the number of existing product categories is extremely large, spanning thousands of values. Moreover, international marketplaces face additional challenges, such as language and dialect diversity and cultural differences, influencing the interpretation of the query. In this work we focus on Q2PT prediction in the global multilocale e-commerce markets. The common approach of training Q2PT models for each locale separately shows significant performance drops in low-resource stores. Moreover, this method does not allow for a smooth expansion to a new country, requiring to collect the data and train a new locale-specific Q2PT model from scratch. To tackle this, we propose to use transfer learning from the highresource to the low-resource locales, to achieve global parity of Q2PT performance. We benchmark the per-locale Q2PT model against the unified one, which shares the training data and model structure across all worldwide stores. Additionally, we compare locale-aware and locale-agnostic Q2PT models, showing the task dependency on the country-specific traits. We conduct extensive quantiative and qualitative analysis of Q2PT models on the large-scale e-commerce dataset across 20 worldwide locales, which shows that unified locale-aware Q2PT model has superior performance over the alternatives.

Figures (5)

• Figure 1: Query distribution for product types worldwide versus an emerging store on an experimental data sample including 20 locales and 1414 product types.
• Figure 2: Precision-recall curves in US (left) and PL (right) locales.
• Figure 3: The query "masque de nuit" exists in both training sets for CA and FR. In CA, the most clicked PT is SLEEPING_MASK while in FR, SAFETY_MASK and SLEEPING_MASK have equal rates, which results in EMD of 0.3
• Figure 4: Density of EMD distances for queries in the intersection of FR $\leftrightarrow$ CA and SA $\leftrightarrow$ EG pairs.
• Figure 5: Examples of product type discrepancies between FR (left 2 products) and CA (right 2 products) stores.