Validating the representation of distance between infarctdiseases using Word2Vec word embedding
Daiki Yokokawa, Kazutaka Noda, Yasutaka Yanagita, Takanori Uehara, Yoshiyuki Ohira, Kiyoshi Shikino, Tomoko Tsukamoto, Masatomi Ikusaka
The paper evaluates Word2Vec embeddings for infarction-related diseases, finding that cosine distance best aligns with ICD-10 codes while Euclidean distance reflects word frequency.
bioRxiv · Jun 21, 2022 · PreprintBioinformatics