In a previous blog post (opens in new tab) , we explained the approach Microsoft Academic takes to the problem of entity linking, including the problem of author disambiguation and conflation. As we continue to push the boundaries of these technologies, we are rolling out a new process that will result in visible changes to some author profiles.
We will explain the technical aspects of the process in a different blog post. For now, one important part of the new process that you need to know about is that Microsoft Academic will use author disambiguation and conflation technologies to verify authorship claims.
So far, a user could claim several authors and/or several papers and assemble them under one profile at will. While most users claimed the papers they themselves had authored, we have discovered that some users claimed papers that were clearly not written by the same individual. We do not assume malicious or dishonest intent, but we would like to ensure that author profiles are accurate.
The question becomes, how can we tell whether a paper or author claimed by a user was indeed written by that user? We have trained our artificial intelligence algorithms to help us with this problem, and starting today, you will see that our AI has cleaned up some user accounts with the intent of increasing accuracy. Specifically, you are likely to see changes around two scenarios:
Scenario 1: Our AI algorithms validate a user’s authorship claim
If our AI agrees with a user’s authorship claim, it will merge the claimed authors into a single author entity and remove old claims. So, when you log into your account and check claimed authors, you are likely to see all your papers listed under only one author.
Furthermore, if our AI finds a paper written by a claimed author, it will automatically add the paper to that user account.
Scenario 2: Our AI algorithms cannot validate a user’s authorship claim
If, according to our algorithms’ best estimate, you are unlikely to be an author of a paper you have claimed, your claims will be removed. This is particularly likely to happen for user accounts where the user’s first and last name are significantly different from claimed author names. If you believe this is a mistake, please contact us by using the Feedback form.
Users who have more than one account with Microsoft Academic may see that all their paper claims have been moved under one of their accounts. We encourage users to maintain only one account.
So, please keep in mind that from now on, Microsoft Academic will verify paper and author claims before honoring them. If claims cannot be corroborated against publicly available information, they will be rejected, and the user will receive a notification.
Claim verification will be reflected on the website Microsoft Academic (opens in new tab), and in the Microsoft Academic Graph (opens in new tab), which you can access via the Academic Knowledge API (opens in new tab) or Azure Data Lake Store (opens in new tab) (please contact us for the latter option).
As a user, there are a few things you can do to help maintain accuracy on Microsoft Academic:
• If your authorship claim is valid and the algorithm made a mistake by rejecting it, we ask for your help to train our artificial intelligence. Please use the feedback form to point us to publicly available online sources that contain correct lists of your publications. Our AI will read them, learn from them, and improve its capability to match papers to their rightful authors.
• Please check the first and last names you provided in your Microsoft Academic profile and ensure they match the name you use on your publications. Avoid using nicknames on your profile. If you use your legal name on your publications (e.g. William) but you give Microsoft Academic your nickname (e.g. Bill), that will make it more difficult for our AI to identify you correctly We are aware that many individuals publish under different names (e.g. legal name change, or some publications include middle initials and others do not). Our AI is able to handle these cases and is continuously learning.
• Please only claim papers you have authored.
We recognize that machine learning is still not perfect, but the scale of the data that we have to cope with has exceeded the capacity of human labor. As a research group, our mission is to push the boundary of AI technologies so that we can harness the power of machines to empower and enhance human ingenuity. Our AI is only 2 years old and is still learning, but as reported in the literature (opens in new tab), the research community seems to agree we are on the right track.
Data accuracy is very important to us and to our users, and we hope that this change in the claim process will improve the experience for all users of Microsoft Academic.
As always, we would like to hear from you either through the feedback link at the bottom right of the website (opens in new tab), or on Twitter (opens in new tab). You can also find our project home page with this blog on the Microsoft Research site at aka.ms/msracad (opens in new tab).
Happy researching!