Structure and Dynamics of Signed Citation Networks

by Srijan Kumar

All citations made in a paper are not equal. To understand the true value of citations, it is important to understand how papers are cited. Therefore, we use the text around the citation to classify the sentiment of the citation into endorsement or positive, criticism or negative, neutral. We make many interesting observations.

#1. Papers citations follow power-law distribution.


#2. Citation networks follow weak balance and status theories.


#3. Authors rarely reciprocate sentiment.

#4. First opinion is the average opinion.


Download the PDF

Download the poster


Citations are important to track and understand the evolution of human knowledge. At the same time, it is widely accepted that all the citations made in a paper are not equal. However, there is no thorough understanding of how citations are created that explicitly criticize or endorse others. In this paper, we do a detailed study of such citations made within the NLP community by differentiating citations into endorsement (positive), criticism (negative) and neutral categories. We analyse this signed network created between papers and between authors for the first time from a social networks perspective. We make many observations – we find that the citations follow a heavy-tailed distribution and they are created in a way that follows weak balance theory and status theories. Moreover, we find that authors do not change their opinion towards others over time and rarely reciprocate the opinion that they receive. Overall, the paper builds the understanding of the structure and dynamics of positive, negative and neutral citations.


@inproceedings{kumar2016structure, author = {Kumar, Srijan}, title = {Structure and Dynamics of Signed Citation Networks}, booktitle = {Proceedings of the 25th World Wide Web Conference companion}, year = {2016}, }


The extracted signed paper citation network and author citation network are here:
Download DATA

The code to extract the networks are here:
Download CODE