Persona-conditioned LLMs improve e-commerce search query generation · arXiv paper

A new arXiv paper explores the effectiveness of persona-conditioned synthetic query generation for e-commerce search, particularly in cold-start scenarios. Researchers generated over 118,000 synthetic query pairs using six LLMs and nine persona-conditioned prompt families on a UK automotive schema. The study found that using the IPIP personality bundle improved parser F1 scores, while syntax conditioning was key to aligning generated queries with real user traffic and increasing lexical diversity. The paper also introduces alignment sKL as a proxy for parser F1 and identifies that dropped constraints are a primary cause of hallucinated queries, with syntax-bearing families showing higher hallucination rates. AI

IMPACT This research could lead to more effective AI-powered search systems by improving the quality and relevance of generated training data.

RANK_REASON The cluster contains an academic paper published on arXiv detailing a new method for synthetic query generation.

Read on arXiv cs.IR (Information Retrieval) →

• alphaXiv
• arXiv
• CatalyzeX
• CORE Recommender
• DagsHub
• Filippos Ventirozos
• Gotit.pub
• Holm-corrected
• Hugging Face
• Influence Flower
• International Personality Item Pool
• ScienceCast
• UK
• von Mises-Fisher distribution

• paper
• product

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →