Research:Detox/Resources
Detox Specific Resources[edit]
- Detox project page
- Detox Data Set
- Detox research paper
- Wikimania 2016 Discussion
- Research Showcase presentation
Relevant Wikipedia Policies[edit]
Community Discussion/ Proposals on Talk Page Abuse and Toxicity[edit]
- Village Pump Proposal
- Community Wish List
- Wikimedia-l discussion
- Revision Scoring Talk Page
- Harassment Consultation
WMF Projects / Discussion on Harassment[edit]
Data Sources[edit]
Wikimedia Sources[edit]
- Administrators Board
- Users Blocked for Harassment (query)
- Edit Filter (about, rules e.g. 294, 478)
- deleted or suppressed talk page comments (very promising)
- Wikilabels (very promising)
External Publicly Available Sources[edit]
- Stanford Politeness Corpus
- MPQA A corpus with annotations for private states (e.g. beliefs, emotions, sentiments, speculations, etc.)
- Kaggle competition to detect insults (very promising)
- Internet Argument Corpus A set of 390,704 posts in 11,800 discussions extracted from the online debate site 4forums.com. Includes: degrees of agreement with a previous post, cordiality, audiencedirection, combativeness, assertiveness, emotionality of argumentation, and sarcasm.
- Alignment and Authority in Wikipedia Discussions (AAWD) Corpus A set of English, Russian, and Mandarin Wikipedia talkpage threads annotated for agreement/disagreement and other social cues.
Other Potential Sources[edit]
- Contact "League of Legends" team for training corpus
- Crowd-source using CrowdFlower, Mechanical Turk, or similar.
Related Work[edit]
- Abuse in Online Games
- A Computational Approach to Politeness
- A Sentiment Analysis Approach for Online Dispute Detection
- Antisocial Behavior in Online Discussion Communities
- How Community Feedback Shapes User Behavior
- Online Harassment Resource Guide
- Like trainer, like bot? Inheritance of bias in algorithmic content moderation (research with the data from the Detox project)
Talk Page Parsing Utilities[edit]
For distilled notes on some of the above resources, click here.