{"slug":"bodyguard-builds-a-machine-learning-model-on-ovhcloud-automl-to-detect-hateful-comments-reaching-90-detection-with-a-3-false-positive-rate","url":"https://findausecase.com/use-cases/bodyguard-builds-a-machine-learning-model-on-ovhcloud-automl-to-detect-hateful-comments-reaching-90-detection-with-a-3-false-positive-rate","title":"Bodyguard builds a machine-learning model on OVHcloud AutoML to detect hateful comments, reaching 90% detection with a 3% false-positive rate","description":"Bodyguard, an anti-cyber-bullying app, built its predictive model and a meta-learning model with OVHcloud AutoML and scikit-learn, running on OVHcloud Public Cloud instances. The source states detection accuracy rose from 80% to 90%, false positives fell from 6% to 3%, and more than 2 million hateful comments were deleted in 20 months.","company":"Bodyguard","industry":"Technology & Software","aiCapabilities":["Predictive Analytics"],"technology":["OVHcloud AutoML","scikit-learn"],"deployment":"Unknown","problemStatement":"Bodyguard's technology needed to analyse the context of a comment and who it is aimed at, understand states of mind, and detect irony, sarcasm and humour, in order to reduce false positives and increase accuracy, with an error rate that could not exceed 10%.","solutionApproach":"Bodyguard used OVHcloud AutoML, a distributed and scalable machine learning platform, together with open-source scikit-learn algorithms to build a predictive model and a meta-learning model that analyses the relationship between the author of the content and the commenter. The infrastructure runs on three OVHcloud Public Cloud instances.","businessValue":"According to the source, detection accuracy rose from 80% to 90% and false positives fell from 6% to 3%. The app deleted more than 2 million hateful comments in 20 months and had more than 40,000 users in July 2019. These figures are the company's own claims.","evidence":{"band":"high"},"sourceUrl":"https://www.ovhcloud.com/en/case-studies/bodyguard/","dates":{"publishedAt":"2026-10-15T05:46:36.785Z","publishedAtSource":"pipeline","updatedAt":"2026-10-15T05:46:36.785Z"},"license":"Open for reading and citing with a link to https://findausecase.com/use-cases/bodyguard-builds-a-machine-learning-model-on-ovhcloud-automl-to-detect-hateful-comments-reaching-90-detection-with-a-3-false-positive-rate. Bulk republication requires permission."}