Diting: An Author Disambiguation Method Based on Network Representation Learning

Abstract

It is important to disambiguate names among persons in many scenarios. In this work, we propose an unsupervised method Diting and a semi-supervised method Diting++ for author disambiguation. In Diting, we learn a low-dimensional vector to represent each paper in networks, which are formed by connecting papers with multiple types of relationship (such as co-author). During representation learning, we focus on maximizing the gap between positive edges and negative edges. Further, we propose a clustering algorithm which associates papers to their real-life authors. To make full use of the authorship information, which is easy to obtain from the authors’ homepages, we design Diting++ to improve the performance for name disambiguation. Diting++ uses the authorship information listed on the authors’ homepages to construct label networks and uses a network representation learning method to learn paper representations based on label networks and other networks. Further, Diting++ uses a semi-supervised clustering method to partition learned paper representations into disjoint groups. Each group belongs to a distinct author. By making use of the label information, the clustering method partitions papers written by the same author in the same group, whereas papers written by different authors locate in different groups. Through extensive experiments, we show that our methods are significantly better than the state-of-the-art author disambiguation methods.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 7 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.