Perspective API
Toxicity Scores & Embeddings
Search and explore comments with their Perspective API toxicity/prosocial scores alongside AI sentiment labels.
|
Toxicity & prosocial scores, embeddings, and clusters generated via
Communalytic
(Social Media Lab, Toronto Metropolitan University) using Google's Perspective API.
Toxicity Scored
55,769
9.3% of 596,542 total
Prosocial Scored
54,229
Embeddings
55,418
403 clusters
Avg Tox / Con
0.245
/
0.328
Summary Charts
click to expand
All 13 Dimensions
Score Distribution
Scored: 55,769
Unscored: 596,542 remaining
9.3% complete
Perspective API Dimensions Reference
13 dimensions explained
Toxic (6)
- Toxicity — Rude, disrespectful, or unreasonable
- Severe Toxicity — Very hateful or aggressive
- Identity Attack — Targeting race, religion, gender, etc.
- Insult — Inflammatory or provocative language
- Profanity — Swear words or obscene language
- Threat — Intention to inflict pain or violence
Prosocial (7)
- Affinity — Agreement or shared understanding
- Compassion — Concern for others' wellbeing
- Curiosity — Desire to learn or understand more
- Nuance — Acknowledges complexity or multiple perspectives
- Personal Story — Shares personal experience
- Reasoning — Evidence-based or logical argumentation
- Respect — Politeness and consideration for others
Data sources:
comment_perspective_scores, comment_embeddings, and view_comment_sentiment ·
Scores are probability values (0–1) from Google's Perspective API via
Communalytic.
Comment Explorer
Browse comments with toxicity & constructive scores. Filter by keyword, polarity, toxicity range, or cluster.