Search Personalization Using Machine Learning
Problem
Optimally ranking search results on websites using consumers' search and click history.
Why it matters
Better ranking helps consumers find products and information, affecting firm performance.
What the AI did
A machine-learning framework (feature generation, nDCG-based LambdaMART, feature selection) personalized rankings using user search and click history, tested via counterfactual analyses on large-scale search engine data.
Results
Personalization improves clicks to the top position by 3.5% and reduces average error in rank of a click by 9.43% over baseline.
Impact
Counterfactual analyses on large-scale data show measurable ranking improvements from personalization.
Record
| Venue | Management Science |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Hema Yoganarasimhan |
| Citations | 167 (OpenAlex, at import time) |
| References | 112 |
| Model or method | LambdaMART (nDCG-based) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1287/mnsc.2018.3255 |
| Open access | green |
| Keywords | search personalization, machine learning, LambdaMART, ranking, click history |
Importance rating
significant — thousands to millions affected
Personalized ranking tested on a leading search engine's large-scale data with measured gains (3.5% top-click lift); benefits consumers and firms in online search.