Writing instructions to get your data annotated by a third-party is a lot trickier than most people think: you need to think of including the right corner cases, and to educate people you don’t know on a use case they might have never worked on before. Should a model car, or a car wreck, count as a car? Should really small objects be annotated, and how small is too small? The ML-Driven Labeling Instructions Generator will guide you through the process and ensure best practices are met and that your labeling instructions are optimized to your dataset, so that you don’t have to fret about those questions anymore.
Information Density-Based Example Selection
A common mistake people make when creating labeling instructions is to use stock images or documents pulled out the internet to define a class. Using data from your own data is one of the easiest ways to dramatically improve the efficacy of your instructions. The Labeling instructions Generator goes one step further and uses ML to identify interesting records that illustrate the challenges of your use case the best.
Quiz-Based Rule Generation
Writing instructions is difficult because it is easy to miss a critical detail that would come across as obvious to a person working on the use case, but not obvious at all to someone seeing the dataset for the first time. The Labeling Instructions Generator solved those challenges via a quiz-like implementation that forces the user to provide the right, unbiased information, and to provide the right examples and counterexamples. It’s easy, and it’s fun, so just give it a try!
Labeling Instructions Strength Measurement
Sending poorly made labeling instructions to your labeling providers can have dire consequences: you might have to iterate countless times with the annotators before they get the full understanding of what they are expected to do, and that might end up costing a lot of time… and a lot of money. The Labeling Instructions Generator module comes equipped with a real-time instructions strength measurement feature which informs you when your instructions are sufficiently polished to be sent.
Labeling Instructions Version Control
Whenever you upload your own or generate labeling instructions with the Alectio Labeling Instructions Generator, those instructions will get properly stored and versions so that you can retrieve them, reuse them, update them, and review what instructions were used for what job or version.
The Alectio system constantly analyzes the mistakes made by the annotators when labeling your data and learns how to improve them, providing you with valuable feedback to improve the efficiency of your future labeling jobs.