Abstract: 【Purpose/significance】Aiming to solve the problem of recognize the authors from same names, and promote the accuracy of
author name disambiguation.【Method/process】This paper proposes to deep express the text features from title, keywords, abstract,
reference, co-authors, email, and affiliation information, utilizing Doc2vec, and then to train and test nine SVM models based on the
features from Doc2vec. Taking PubMed Central data as an example, this paper disambiguates all author once the model has higher ac?
curacy in test.【Result/conclusion】As results shown, this method can effectively promote the accuracy of author recognition, up to
91.80%. Our method not only takes full use of the features from affiliation, email and co-author, but also follows the succession and
evolution of research content, integrates multiple features to solve the problem of author name disambiguation, especially for no affilia?
tion and email, which shows more potential application when authors flow more frequently.【Innovation/limitation】The integration
method of unsupervised text representation learning and supervised machine learning, show good performance for the author data from biomedical and life science