Reproduce “Bridging Personalization and Control in Scientific Personalized Search” 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
Sheshera Mysore, Garima Dhanania, Kishor Patil, Surya Kallumadi, Andrew McCallum, Hamed Zamani
CtrlCE, developed by researchers at the University of Massachusetts Amherst, presents a memory-augmented cross-encoder designed for scientific personalized search that integrates user control. The model achieves competitive retrieval performance while allowing users to actively influence search results through transparent, editable user profiles and intelligently identifies when user intervention is most beneficial.
Key insights
- Integrating an embedding cross-encoder with multi-vector user memories allows for high-performance personalized search that also supports transparency and user agency.
- A calibrated mixing model can intelligently determine when personalization is beneficial or when user input is most needed, reducing unnecessary user burden.
- Designing user profiles as editable concept-value pairs or item collections enables users to directly influence personalized search results and mitigate 'filter bubble' effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.