Topic-driven multi-type citation network analysis

Abstract

In every scientific field, automated citation analysis enables the estimation of importance or reputation of publications and authors. In this paper, we focus on the task of ranking authors. Although previous work has used content-based approaches or citation network link analyses, the combination of the two with topical link analyses is unexplored. Moreover, previous citation analysis applications are typically limited to a graph based on author citations, or a bipartite graph based on author and paper citations. We present in this paper a novel integrated probabilistic model which combines a content-based approach with a multi-type citation network which integrates citations among papers, authors, affiliations and publishing venues in a single model. We further introduce the application of Topical PageRank into citation network link analysis due to the fact that researchers may be experts in different scientific domains. Finally, we describe a heterogenous link analysis of the citation network, exploring the impact of weighting various factors. Comparative experimental results based on data extracted from the ACM digital library show that 1) the multi-type citation graph works better than citation graphs integrating fewer types of entities, 2) the use of Topical PageRank can further improve performance, and 3) Heterogenous PageRank with parameter tuning can work even better than Topical

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 — 20 citations exist upstream.

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