As described in our March 23, 2018 post, we unveiled a feature at the Microsoft Academic website that invites users to claim authorship of their publications. This feature is particularly crucial to authors with common names, as we have deliberately taken a conservative stance in author disambiguation. It is not uncommon for many of you to find that we have assigned your publications into multiple author clusters because, from our machine’s perspective, your co-authors, affiliations, and topics of those publications vary significantly. Conversely, if you see we recognize many publications as from a single author, chances are our machine has found multiple examples of convincing evidence to support that decision.
Since this feature became available a year ago, there have been thousands of requests per week regarding authorship claims. These authorship claims have dramatically helped the twice-monthly updated Microsoft Academic Graph (MAG) in reducing its under-conflation issues, as described in a previous blog post . While the overwhelming majority of these requests are legitimate, we have seen some undesirable behaviors on our website. Given the site’s search capabilities, any false claims will be immediately visible to the public and rebuked. Nevertheless, knowing mistakes (and fraud) can still happen, we have designed an automated claim verification to weed out claims by obviously incompatible authors.
We did not expect that anyone would try to pretend to be another person, like the example below where a person called “Vũ Khoa Nguyễn” claimed authorship of all the papers by Yoshua Bengio (which has since been corrected using the mechanism described below).
The irony is, the more famous an author pretends to be, the easier we can detect such illicit behavior. With a few keystrokes and clicks, one can easily verify that the famous AI professor Yoshua Bengio does not have a Vietnamese alias. Claiming all his publications under a different name fools no one and serves little purpose other than exposing the claimant’s falsehoods. To be sure, the person whose name is used in the authorship claim may well be an identity theft victim themselves. Given that the identity services we use cannot be foolproof, such false claims are bound to happen. We would like to emphasize that we have designed our automated verification system to defend the integrity of MAG against this scenario. In other words, the fake claims are excluded from MAG distribution and have no effect on any analytics derived from the graph. Only the Microsoft Academic website is impacted.
Having said that, we do understand such vandalism creates falsehoods. As we strive to provide the best user experience, we are introducing a new measure to scrub these inaccuracies from our website. Starting in April 2019, we will compare the name of the author being claimed with that that we get through the identity service. The claims will be rejected if the difference is significant. If the name you give to your identity service provider is already the name you publish under, you are all set. However, if you have published under various names, we kindly request you use the feedback button on our website to let us know your situation, once you receive a claim rejection from our automated system. We will program the exception into our system for you to reclaim again, and then update the MAG and the website in the following cycle. Please note that each exception is personal and only applicable to you. Any others, such as your siblings or domestic partners who have the same name mismatch case will have to inform us separately. Our goal is that this new name-check policy offers the best balance between accuracy and user experience.
Happy researching!