Additional criteria for various types of tasks
Field tasks and selecting areas in images
Field tasks are completed in the Toloka mobile app. Usually the performer should go to a specified address, take photos or record a video, and answer certain questions. In tasks for selecting image areas, the performer should select an area by drawing a polygon or rectangle in an image.
- Checking completed tasks
These types of tasks usually have non-automatic acceptance, so the performers' responses need to be checked. There are three ways to do this:
Auto checking. For example, you wrote a script that retrieves the results via the API, sends the responses for recognition and review, checks the coordinates, and then returns the task status (accept or reject), and uploads them to Yandex.Toloka. If you created a script that does this or you automated the review in another way, show us how it works.
Review task. You can assign other performers a Yandex.Toloka task for reviewing completed assignments. It's important that the review task isn't available to performers who participated in the field task.
- Task acceptance
We expect your project to be configured as follows:
The average task review period is no more than 7 calendar days. Preferably it should be 2-3 days.
The task acceptance rate is at least 75%. Our experience shows that a high percentage of rejected tasks is an indicator of poor process management and this should be fixed. Most often, a performer is assigned a very large and complex task that can and should be divided into several simple ones.
Classification is a multiple choice task. Examples are moderating comments or grouping images by category.
- Quality control
Rules help you measure the quality of performers' responses in real time: set a higher price for those who do well and restrict access to tasks for cheaters.
Use at least three different quality control rules in the project, one of which must be Control tasks.
- None of us like to read long instructions. It's much easier to learn something from examples. Training helps users figure out how to do the task and you can choose the ones who are good at it. Update your training pool at least once every three months. Otherwise, it becomes outdated and performers will be able to learn it by heart.