Our paper on extracting drug-drug interaction (DDI) from the literature has been accepted for publication in Bioinformatics. This paper focuses on using heterogeneous drug-related domain information for DDI relation extraction. Especially, we have created a large-scale knowledge graph covering drug-related information and learned embeddings of the nodes and links in order to obtain embedding vectors of various pieces of domain information in a single vector space, which is very important. Then, we use the pretrained embedding representations as external knowledge for neural DDI extraction from the literature.
This is a typical example of symbolic-neural learning (or neuro-symbolic learning), which we have been promoting in a series of International Workshops on Symbolic-Neural Learning since 2017.
Masaki Asada , Makoto Miwa and Yutaka Sasaki, Integrating heterogeneous knowledge graphs into drug-drug interaction extraction from the literature, Bioinformatics, Oxford University Press, November 23, 2022. (https://doi.org/10.1093/bioinformatics/btac754)
Computational Intelligence Laboratory, TTI, Japan 