报告题目: Ontology and Word Embedding for Biomedical Language Processing
摘要：Current natural language processing methods heavily rely on vector space word representations, known as word embeddings, self-trained on large text corpora. Knowledge graphs are also increasingly represented in vector spaces as graph embeddings. Ontologies are instances of knowledge graphs. I will present a short review of the methods created to embed words and ontologies. I will then focus on a method that we designed to create a shared embedding space for both corpus words and ontology concepts. It is applied to a concept normalization task, which consists in mapping an input term to the most relevant concept in a target ontology. This concept normalization task is a part of the Bacteria Biotope BioNLP shared tasks, for which our method holds a top rank.