Today, many news sites let users write comments on news articles, rate others' comments by upvoting and downvoting, and order the comments by the rating. Top-rated comments are placed right below the news article and read widely, reaching a large audience and wielding great influence. As their importance grew, upvotes and downvotes are increasingly manipulated by coordinated efforts to hide existing top comments and push certain comments to the top. In this paper, we analyze comment sections of articles targeted by coordinated efforts and identify a trace of vote manipulation. Based on the findings, we propose a parameterized classifier that distinguishes comment threads affected by coordinated voting. The classifier only uses the number of upvotes and downvotes of comments. Therefore it is widely applicable to general vote-based curation systems where contents are sorted by the difference of upvotes and downvotes. Using the classifier and our choice of parameters, we have examined six years of the entire commenting history on a leading news portal in South Korea. Manual inspection in partisan online communities could only identify a few hundreds of targeted articles. With our classifier, we have identified more than ten thousand comment threads with a high likelihood of manipulation. We also observe a significant increase in coordinated manipulation in recent years.
|Title of host publication||Proceedings of the 14th International AAAI Conference on Web and Social Media, ICWSM 2020|
|Number of pages||12|
|Publication status||Published - 2020|
|Event||14th International AAAI Conference on Web and Social Media, ICWSM 2020 - Atlanta, Virtual, United States|
Duration: 2020 Jun 8 → 2020 Jun 11
|Name||Proceedings of the 14th International AAAI Conference on Web and Social Media, ICWSM 2020|
|Conference||14th International AAAI Conference on Web and Social Media, ICWSM 2020|
|Period||20/6/8 → 20/6/11|
Bibliographical noteFunding Information:
This work was partially supported by Barun ICT Research Center at Yonsei University and the Institute for Basic Science (IBS-R029-C2). The data collection and interpretation was possible because Naver News archives all the previous articles and comments, and provides their policy modifications to the public. We thank Naver News for their data availability and open policy.
Copyright © 2020, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.
All Science Journal Classification (ASJC) codes
- Computer Networks and Communications