TY - GEN
T1 - Context-aware detection of sneaky vandalism on wikipedia across multiple languages
AU - Tran, Khoi Nguyen
AU - Christen, Peter
AU - Sanner, Scott
AU - Xie, Lexing
N1 - Publisher Copyright:
© Springer International Publishing Switzerland 2015.
PY - 2015
Y1 - 2015
N2 - The malicious modification of articles, termed vandalism, is a serious problem for open access encyclopedias such as Wikipedia. Wikipedia’s counter-vandalism bots and past vandalism detection research have greatly reduced the exposure and damage of common and obvious types of vandalism. However, there remains increasingly more sneaky types of vandalism that are clearly out of context of the sentence or article. In this paper, we propose a novel context-aware and cross-language vandalism detection technique that scales to the size of the full Wikipedia and extends the types of vandalism detectable beyond past feature-based approaches. Our technique uses word dependencies to identify vandal words in sentences by combining part-of-speech tagging with a conditional random fields classifier. We evaluate our technique on two Wikipedia data sets: the PAN data sets with over 62, 000 edits, commonly used by related research; and our own vandalism repairs data sets with over 500 million edits of over 9 million articles from five languages. As a comparison, we implement a feature-based classifier to analyse the quality of each classification technique and the trade-offs of each type of classifier. Our results show how context-aware detection techniques can become a new counter-vandalism tool for Wikipedia that complements current feature-based techniques.
AB - The malicious modification of articles, termed vandalism, is a serious problem for open access encyclopedias such as Wikipedia. Wikipedia’s counter-vandalism bots and past vandalism detection research have greatly reduced the exposure and damage of common and obvious types of vandalism. However, there remains increasingly more sneaky types of vandalism that are clearly out of context of the sentence or article. In this paper, we propose a novel context-aware and cross-language vandalism detection technique that scales to the size of the full Wikipedia and extends the types of vandalism detectable beyond past feature-based approaches. Our technique uses word dependencies to identify vandal words in sentences by combining part-of-speech tagging with a conditional random fields classifier. We evaluate our technique on two Wikipedia data sets: the PAN data sets with over 62, 000 edits, commonly used by related research; and our own vandalism repairs data sets with over 500 million edits of over 9 million articles from five languages. As a comparison, we implement a feature-based classifier to analyse the quality of each classification technique and the trade-offs of each type of classifier. Our results show how context-aware detection techniques can become a new counter-vandalism tool for Wikipedia that complements current feature-based techniques.
UR - http://www.scopus.com/inward/record.url?scp=84945918241&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-18038-0_30
DO - 10.1007/978-3-319-18038-0_30
M3 - Conference contribution
SN - 9783319180373
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 380
EP - 391
BT - Advances in Knowledge Discovery and Data Mining - 19th Pacific-Asia Conference, PAKDD 2015, Proceedings
A2 - Ho, Tu-Bao
A2 - Motoda, Hiroshi
A2 - Motoda, Hiroshi
A2 - Lim, Ee-Peng
A2 - Cao, Tru
A2 - Cheung, David
A2 - Zhou, Zhi-Hua
PB - Springer Verlag
T2 - 19th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2015
Y2 - 19 May 2015 through 22 May 2015
ER -