“…Neural architectures have been widely used in recent state-of-the-art models for MCN from scientific texts, user reviews, social media texts and clinical notes ( Ji et al , 2020 ; Leaman and Lu, 2016 ; Li et al , 2017 , 2019 ; Miftahutdinov and Tutubalina, 2019 ; Sung et al , 2020 ; Xu et al , 2020 ; Zhao et al , 2019 ; Zhu et al , 2020 ). Most models share limitations regarding a supervised classification framework: (i) to retrieve concepts from a particular terminology for a given entity mention, models are required re-training, (ii) use additional classification or ranking layer, therefore, during inference compute all similarities between a given mention and all concept names from a dictionary through this layer and sort these scores in descending order.…”