The second example is about my advisor, Dr. Dongwon Lee’s publications in DBLP. Because “Dongwon Lee” is one of common names, there are another author with the same name in digital libraries. In addition, DBLP does not have any unique identifier corresponding to each author name. Therefore, the citations by another scholar with the same name spelling can be merged into a single collection. In fact, citation 22 and citation 24 really belong to another “Dongwon Lee” at University of Minnesota.

This figure shows the accuracy of both approaches for EconPapers. For example, when there is a single false citation hidden in the 100 citations, the sampling approach with the bag model can identify the false citation with more than 60% accuracy in the bottom of 10%, more than 80% accuracy in the bottom of 20%, etc.