Improving author name disambiguation with user relevance feedback.

Abstract

Author name ambiguity in the context of bibliographic citations is a very hard problem. It occurs when there are citation records of a same author under distinct names or when there exists citation records belonging to distinct authors with very similar names. Among the several methods proposed in the literature, the most effective ones are those that perform a direct assignment of the records to their respective authors by means of the application of supervised machine learning techniques. However, those methods usually need large amounts of labeled training examples to properly disambiguate the author names. To deal with this issue, in previous work, we have proposed a method that automatically obtains and labels the training examples, showing competitive performance compared to representative author name disambiguation methods. In this work, we propose to improve our previous method by exploiting user relevance feedback. In more details we select a very small portion of the citation records for which our method was mostly unsure about the correct authorship and ask the administrators for labeling them. This feedback is then used to improve the effectiveness of the whole process. In our experimental evaluation, we observed that with a very small labeling effort (usually around 5% of the records), the overall disambiguation effectiveness improves by almost 10% on average, with gains of up to 61% in some of the largest ambiguous groups.