Sequence-to-sequence models optimized for better query generation in e-commerce search
Query Generation with Direct Preference Optimization for Document Expansion in E-commerce Search
Information Retrieval
Summary
Search engines sometimes struggle to find products because the words people use don’t match those in product descriptions. To fix this, the authors improved a method that generates search queries based on product documents to help match user searches better. They trained their system to prefer more relevant and new queries while filtering out irrelevant or repetitive ones. This led to fewer wrong or repetitive search suggestions and improved user experience on a large shopping website.
What this means in practice
- •For e-commerce search engineers: Improve product search accuracy by generating more relevant query expansions that reduce irrelevant and repetitive matches in product indexing.
- •For online retail platform operators: Increase user engagement and search relevance by deploying optimized query generation methods for product documents in live e-commerce search systems.$Commercial implications: Enables retail platforms to offer superior search experience, attracting more users and conversions, thereby generating commercial advantage.
Authors
Kaihao Li, Feng Liu, Juexin Lin, Xunfan Cai, Zhen Yang, Tony Lee, Ciya Liao
Abstract
Doc2Query, a popular document expansion technique, leverages sequence-to-sequence models to generate relevant queries, effectively addressing the "vocabulary mismatch" problem in information retrieval. However, these models often suffer from generating either hallucinations unrelated to the document or repetitive content already present in the document. Training sequence-to-sequence models to produce high-quality, novel, and relevant tokens remains a significant challenge. To address these issues, we introduce a novel approach, QGDPO, that employs direct preference optimization (DPO) to guide the generation process. We first fine-tune a base sequence-to-sequence model and subsequently utilize a relevance model to score its predictions. Based on these scores, we construct pairs of winning and losing predictions as relevance preferences for the DPO training. Furthermore, we enhance our pipeline by using the relevance model to filter out poor predictions, retaining only the most relevant generated content for indexing. QGDPO effectively eliminates 50% of irrelevant predictions comparing against Doc2Query baselines, while the relevance filter removes an additional 14.61%. This feature has been successfully deployed to production on Walmart.com for full traffic, with a substantial improvement in relevance and user engagement.