Accepted for JBI

Another full paper has been accepted for publication in the Journal of Biomedical Informatics.

Tomoki Tsujimura, Makoto Miwa, Yutaka Sasaki, Large-scale neural biomedical entity linking with layer overwriting, Journal of Biomedical Informatics, Elsevier, 2023. (to appear) (IF=8)

This paper reports a pure neural entity linking model that can link each entity to one of millions of biomedical concept classes. As the number of concepts which are included in a training set is much smaller than the real size of biomedical databases, we augment the training data with terms and synonyms of database entries and conduct two step training. The first step is ordinary training with all the data (i.e. original training data+ augmented data) and then the second step conducts layer overwriting at the final output layer. This trick is not easy to explain in short but it is clearly effective to improve test scores.