Modeling Scientific Impact with Topical Influence Regression

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 113–123, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Modeling Scientific Impact with Topical Influence Regression James Foulds Padhraic Smyth Department of Computer Science University of California, Irvine {jfoulds, smyth}@ics.uci.edu Abstract When reviewing scientific literature, it would be useful to have automatic tools that iden- tify the most influential scientific articles as well as how ideas propagate between articles. In this context, this paper introduces topical influence, a quantitative measure of the ex- tent to which an article tends to spread its topics to the articles that cite it. Given the text of the articles and their citation graph, we show how to learn a probabilistic model to re- cover both the degree of topical influence of each article and the influence relationships be- tween articles. Experimental results on cor- pora from two well-known computer science conferences are used to illustrate and validate the proposed approach. 1 Introduction Scientific articles are not created equal. Some ar- ticles generate entire disciplines or sub-disciplines of research, or revolutionize how we think about a problem, while others contribute relatively little. When we are first introduced to a new area of scien- tific study, it would be useful to automatically find the most important articles, and the relationships of influence between articles. Understanding the im- pact of scientific work is also crucial for hiring deci- sions, allocation of funding, university rankings and other tasks that involve the assessment of scientific merit. If scientific works stand on the shoulders of giants, we would like to be able to find the giants. The importance of a scientific work has previ- ously been measured chiefly through metrics derived from citation counts, such as impact factors. How- ever, citation counts are not the whole story. Many citations are made in passing, are relevant to only one section of an article, or make no impact on a work but are referenced out of “politeness, policy or piety” (Ziman, 1968). In reality, scientific impact has many dimensions. Some articles are important because they describe scientific discoveries that alter our understanding of the world, while some develop essential tools and techniques which facilitate future research. Other articles are influential because they introduce the seeds of new ideas, which in turn in- spire many other articles. In this work we introduce topical influence, a quantitative metric for measuring the latter type of scientific influence, defined in the context of an un- supervised generative model for scientific corpora. The model posits that articles “coerce” the articles that cite them into having similar topical content to them. Thus, articles with higher topical influence have a larger effect on the topics of the articles that cite them. We model this influence mechanism via a regression on the parameters of the Dirichlet prior over topics in an LDA-style topic model. We show how the models can be used to recover meaningful influence scores, both for articles and for specific ci- …

Content truncated to 500 words