Dimensions Adapter for Profiles RNS

Moisey Gruzman, Eric Meeks, Brian Turner

Keywords: machine learning, software API

Abstract:

UCSF Profiles been built using Profiles RNS developed by Harvard University with additions requested by UCSF Faculties. We could extend this application to cover several UC Institutions, trying to satisfy their specific. Most of them could be covered by extending UCSF developed ORNG mechanism. But Profiles RNS has limitations and profile owners complaining about limited set of publications. They are:
- It has PubMed only source of publications;
- It can’t disambiguate publications from PubMed for persons published in multiple colleges;
- It has problem disambiguating Asian Names;
- PubMed has limited animal health publications set;
To resolve those issues and deal with new sets of Data, we've developed machine learning Adapter adding publications from Dimensions Repository into Profiles RNS.
Adapter works gathering information about authors, publications from PubMed, Dimensions and learns how to search publications in both Repositories avoiding duplication. It uses PubMed publications linked to author(s) in Profiles RNS, DOI and ISSN of those publications, names and ORCIDs of authors to find Dimensions author ID to be used in Processor (Filter) to find new publication(s) for that author ID. Preprocessor runs every week and Observer adjusts (Filter) on base of newly found data. We recognized, PubMed as well as Dimensions have errors and lost data, so Observer can found those analyzing existing and newly coming information from Dimensions and Profiles RNS where publication data may be corrected by author manually.
We have passed Stage1 of Adapter into Production and Adapter has added 10% publications into UCSF profiles. Even more been added for UC Davis profiles.Results:
For https://profiles.ucdavis.edu/kwan-liu.ma 245 publications from Dimensions and only 86 from PubMed
For https://profiles.ucsf.edu/mark.vonzastrow 11 publications from Dimensions and only 5 from PubMed
For https://profiles.ucsf.edu/catherine.smith 34 publication from Dimensions and only 31 from Pubmed