OBLIQ-IR: Search relevance based on stance or style
A post sharing the OBLIQ-IR paper describes researchers training a dense retriever—a search model—for queries where relevance depends on traits beyond topic overlap.
TLDR
A post describing OBLIQ-IR says researchers train a dense retriever for “oblique queries,” where relevance depends on latent traits such as stance or style rather than topic overlap. It links to an arXiv paper and a GitHub repository for the project.
Combined views
77
1 Source, first seen ago