Reproduce “Query Rewriting via Cycle-Consistent Translation for E-Commerce Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiming Qiu, Kang Zhang, Han Zhang, Songlin Wang, Sulong Xu, Yun Xiao, Bo Long, Wen-Yun Yang
Researchers at JD.com introduced a cyclic machine translation framework for automated query rewriting in e-commerce search, leveraging Transformer models and a novel cyclic consistent training algorithm on click log data. This approach led to a 0.5219% increase in User Conversion Rate and a 1.1054% increase in Gross Merchandise Value during online A/B tests.
Key insights
- Reformulating query rewriting as a cyclic machine translation problem effectively leverages abundant query-to-title click log data for automated, data-driven transformations.
- A cyclic consistency loss during training significantly improves the quality of rewritten queries by ensuring the sequential translation (query title query) accurately reconstructs the original input.
- Strategic system optimizations, including offline pre-computation, a hybrid Transformer-RNN model for online inference, and syntax tree merging, are crucial for deploying deep learning models in high-traffic industrial e-commerce search.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.