Research & Papers

IntentTune uses personal search history to decode vague e-commerce queries

For under-specified queries like 'watch', search history beats population trends 3x.

Deep Dive

A new paper introduces IntentTune, a framework that tackles one of e-commerce search's toughest problems: under-specified queries. When a user types 'watch' or 'shirt,' the system has no explicit attributes like gender or age group. The researchers compared two approaches: using population-level demand patterns aggregated across all users versus user-specific behavioral signals including search history, browsing activity, and profile attributes.

Their experiments on real-world e-commerce data revealed that population-level patterns alone are insufficient for reliable intent inference. However, user-specific signals—particularly prior search queries—dramatically improved accuracy in predicting gender, age group, product category, and size. This suggests that personalization can resolve the 'unknown intent' problem without requiring explicit demographic input from shoppers.

Key Points
  • Prior search queries from a user's history were the single strongest predictor of intent for vague terms like 'watch' or 'shirt'.
  • Population-level demand patterns alone failed to reliably infer gender, age group, or size for under-specified queries.
  • The framework also leveraged browsing activity and profile attributes, but search history outperformed both by a significant margin.

Why It Matters

Personalization in e-commerce search can now handle vague queries, reducing friction and improving conversion rates.

📬 Get the top 10 AI stories daily