[Paper Review] Clustering of scientific citations in Wikipedia

Finn Årup Nielsen|arXiv (Cornell University)|Jan 1, 2008

Wikis in Education and Collaboration9 citations

TL;DR

This paper proposes using non-negative matrix factorization (NMF) on a Wikipedia (article × journal) citation matrix to perform soft clustering of articles and journals, revealing underlying scientific topics. The key contribution is demonstrating that citation patterns in Wikipedia can effectively uncover thematic groupings in scientific literature, with clusters aligning meaningfully with academic disciplines.

ABSTRACT

The instances of templates in Wikipedia form an interesting data set of structured information. Here I focus on the cite journal template that is primarily used for citation to articles in scientific journals. These citations can be extracted and analyzed: Non-negative matrix factorization is performed on a (article x journal) matrix resulting in a soft clustering of Wikipedia articles and scientific journals, each cluster more or less representing a scientific topic.

Motivation & Objective

• To analyze the structure of scientific citations in Wikipedia using structured template data.
• To identify latent scientific topics by clustering Wikipedia articles and journals based on citation patterns.
• To evaluate whether citation co-occurrence in Wikipedia reflects meaningful scientific disciplines.
• To demonstrate the utility of Wikipedia's citation data as a scalable proxy for scientific knowledge organization.

Proposed method

• Extracting 'cite journal' templates from Wikipedia articles to build a (article × journal) citation matrix.
• Applying non-negative matrix factorization (NMF) to the citation matrix to identify low-rank, non-negative components.
• Interpreting each NMF component as a soft cluster representing a scientific topic, with articles and journals assigned to multiple clusters based on citation weights.
• Using the resulting factorization to group articles and journals into thematic clusters based on shared citation patterns.
• Validating cluster coherence by examining the topical relevance of dominant journals and articles within each cluster.

Experimental results

Research questions

• RQ1 Can citation patterns in Wikipedia be used to identify meaningful scientific topics through clustering?
• RQ2 To what extent do the clusters derived from Wikipedia citations correspond to established scientific disciplines?
• RQ3 How well do the resulting soft clusters represent the thematic content of articles and journals?
• RQ4 What is the effectiveness of NMF in revealing structure in large-scale, real-world citation data from Wikipedia?

Key findings

• The NMF-based clustering successfully identified distinct scientific topics from Wikipedia citation patterns.
• Each cluster contained articles and journals that were thematically coherent, with dominant journals reflecting core publications in specific fields.
• Articles and journals were assigned to multiple clusters, reflecting the interdisciplinary nature of scientific research.
• The method revealed that citation structures in Wikipedia reflect real-world scientific domain boundaries, validating its use as a knowledge discovery tool.