Reproduce “Contextually Aware E-Commerce Product Question Answering using RAG” 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
Praveen Tangarajan, Anand A. Rajasekar, Manish Rathi, Vinay Rao Dandin, Ozan Ersoy
Researchers at the Flipkart US R&D Center developed a contextually aware Retrieval Augmented Generation (RAG) framework for e-commerce Product Question Answering. This system, deployed to over 5 million monthly active users, achieved an 8% increase in user thumbs-up rates and delivered measurable improvements in conversion and customer satisfaction.
Key insights
- A Standalone Query (SAQ) module, fine-tuned on an in-house LLaMA 3–8B model, effectively rephrases conversational queries, achieving 95.93% accuracy in multi-turn scenarios.
- An entropy-based intent model and domain-adapted bi-encoder for retrieval efficiently handle diverse query types and heterogeneous data, with the retrieval module achieving 98.32% Recall@k.
- Strategic LLM prompting combined with responsible "I Don't Know" responses ensures high factual grounding and low hallucination rates by strictly adhering to retrieved context.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.