Case Study: Artimys achieves 4–5× improvements in online bullying detection accuracy with CrowdFlower

Artimys Language Technologies Train Machine Learning Models

Artimys Language Technologies helps parents protect children from online bullying, sexual predators, and signs of suicidal behavior by monitoring real-time online conversations. The company struggled to train machine-learning models because bullying language is highly subjective and rapidly evolving, and it needed large amounts of high-quality, human-labeled ground-truth data to distinguish true bullying from false positives like profanity.

Artimys fed 40,000 high-likelihood snippets (drawn from a 2 million-message corpus) into Figure Eight’s platform, where thousands of workers produced 150,000+ judgments in hours to create accurate labels. The labeled data boosted Artimys’s bully-detection model—precision rose 5.2×, recall improved 25%, and F1 increased 4.2×—yielding 4–5× gains in key measures and enabling more accurate, real-time protection for children online.

Artimys

Bob Dillon

CEO, Artimys