Google labels millions of data points in minutes with Snorkel AI

Google managed hundreds of topic and product classifiers across Search, Ads, YouTube and Analytics, and manual labeling for each new classifier took months, with business changes forcing costly relabeling. Using "Snorkel DryBell," Google engineers wrote labeling functions encoding heuristics and reused internal resources - URL rules, custom NER, and internal topic models for topic classifiers, and keyword matching plus the Knowledge Graph for multilingual product classifiers. The approach completed roughly six months of manual labeling in 30 minutes, delivered a 52% average accuracy improvement, and programmatically labeled 6.5 million data points for product classification and 684,000 for topic classification, matching the quality of tens of thousands of hand-labeled examples; the work was published at SIGMOD 2019.

Snorkel AI is a data-development and evaluation company whose programmatic labeling tools build training and benchmark datasets used to stress-test whether a model approach is viable before a full build.

This sits in the Design & Stress-Test stage of the AI use-case lifecycle: Named examples from the tools pharma teams use to scope, prototype and benchmark an AI use case before committing to build it — data-catalog, sandbox and evaluation platforms.