Abstract: Highlights•The proposal applies dual-branch learning to separate local and global information.•It projects local and global text features into distinct latent space.•Our method enhances performance efficiently without architectural changes.
External IDs:dblp:journals/prl/KooSK25
Loading