EMNLP2025 in Suzhou, China, from November 4th to November 9th, was a blast! Boshko Koloski and Matej Martinc presented SEKE: Specialized Experts for Keyword Extraction—an architecture built on the Mixture-of-Experts idea for keyword extraction with encoder-style models, resulting in improved SOTA and interpretable expert specialisations (ranging from lexical on smaller domains to semantic—named entities—on larger domains such as news). Co-authored with Hanh Thi-Hong Tran and Senja Pollak 🚀
More about the paper, poster, and video presentation here: https://lnkd.in/dNjUKS24
