• Etsy uses supervised ML to detect policy violations across 100M items. • Trust & Safety team combines community reports with automated removal. • Over 100,000 violations flagged and removed in the past year. • Datasets built from annotated positives and hard negatives for accuracy. • ML models trained on real marketplace data to scale detection. • Focus on protecting women‑led creative businesses and fostering trust.

Article Summaries:

  • Etsy’s Trust & Safety team is expanding its content‑moderation capabilities by deploying machine‑learning (ML) models to detect policy violations at scale. Over the past year the system has identified and removed more than 100,000 infringing listings. The approach relies on supervised learning, training on curated datasets that include confirmed violations (positives), hard negatives (near‑misses), and soft negatives (random non‑violating items). Each example is enriched with multimodal signals from text and images. Datasets are split temporally for progressive evaluation, and models-featuring a text encoder and other feature extractors-are tuned on training, validation, and test sets to ensure robust, unbiased performance.
  • Etsy has expanded its Trust & Safety toolkit by deploying a supervised‑learning system to detect policy violations at scale. The model, trained on over 100,000 annotated listings, uses multimodal data-text and images-to predict whether an item breaches marketplace rules. Datasets include hard negatives (near‑violations) and soft negatives (random non‑violations) to improve generalisation, and are split by time for progressive evaluation. Since launch, the system has removed more than 100,000 violations in the past year, helping Etsy keep its 100 million‑item marketplace safe for its 90 million buyers and 7 million sellers.
  • Etsy’s Trust & Safety team has deployed a supervised‑learning system to scale content moderation across its marketplace, which hosts over 100 million listings and serves 90 million buyers. The new model, trained on annotated data that includes both confirmed violations (positives) and non‑violating items (hard and soft negatives), uses multimodal features from text and images. By splitting data chronologically for progressive evaluation, the team fine‑tunes hyper‑parameters and guards against over‑fitting. In the past year the system has identified and removed more than 100,000 policy violations, helping Etsy maintain a safe environment for its largely female‑owned seller base.
  • Etsy has expanded its Trust & Safety operations by deploying a supervised‑learning system to detect policy violations across its marketplace. Leveraging over 100 million listings, the company built annotated datasets of confirmed violations (positives), hard negatives (near‑misses), and easy negatives (random non‑violations) to train a multimodal model that processes both text and imagery. The system, trained with progressive time‑based splits to mimic production, has already identified and removed more than 100,000 violations in the past year. Etsy’s approach emphasizes continuous dataset refinement and model tuning to improve detection accuracy and scalability.
  • Etsy has expanded its Trust & Safety program by deploying a supervised‑learning system to detect policy violations at scale. The company now uses machine‑learning models that analyze both text and images from listings, trained on curated datasets of confirmed violations (positives), hard negatives (near‑violations), and easy negatives (random non‑violating items). Over the past year, the system has identified and removed more than 100,000 infringing items. Etsy’s approach includes progressive evaluation-splitting data by time to mimic production-and a neutral class for non‑violating listings, aiming to improve accuracy and generalizability as the marketplace grows.

Sources: