Reproduce “Improving Item Discoverability in e-Commerce Search via Related Intent Generation” 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
Ji Xin, Xiao Xiao, Ishan Bhatt, Vinesh Gudla, Trace Levinson, Raochuan Fan, Shishir Kumar Prasad, Prakash Putta and 1 more
Instacart researchers developed a hybrid LLM/SLM system for e-commerce search that generates implicit user intents, increasing item discoverability for a broader range of products. This approach expanded query coverage from 60% to 80% and improved purchase prediction F1 scores for high-traffic queries while handling long-tail queries cost-effectively.
Key insights
- Explicitly modeling and generating implicit user intents (substitute, complementary, thematic) significantly enhances item discoverability beyond direct query matches, addressing latent user needs.
- A hybrid LLM/SLM architecture offers a scalable and cost-effective solution for deploying generative AI in real-time e-commerce search, achieving broad query coverage while mitigating operational challenges.
- Teacher-student distillation, augmented with contextual metadata, effectively transfers complex LLM reasoning capabilities to smaller models, enabling high-quality intent generation for the long tail.
- A dual evaluation strategy, combining downstream utility (purchase prediction) with intrinsic semantic quality, is crucial for comprehensively assessing the value of discovery-oriented search systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.