BLOG Machine Learning

What’s the difference between Playment and Traditional Crowdsourcing

Mothi Venkatesh
December 1, 2017

Most of the enterprise companies use the crowd to get work done. Infact, most Playment customers used Mechanical Turk. So how is Playment different?Crowdsourcing certainly has its own merits of flexibility, scalability, and cost-effectiveness. But it fails at  “Quality Assurance” which is a bigger deal than all of them put together.That’s why Playment has been smartly devised to combine merits of crowdsourcing along with guaranteed quality assurance.Salient features that differentiate Playment are,State-of-the-art Tools:

  1. Efficient tools to support various classification, annotations and transcription tasks with custom features for varied task types.
  2. Gamified-interface that amplifies the speed of workers and increases engagement.
  3. Workflow tool to automate and manage a series of complex microtasks with the one-time setup.
  4. Playment’s proprietary algorithms ensure high accuracy level.

High skilled Annotators:

  1. Identities, behaviors, and skills are known.
  2. Well trained with practice tasks for maximum throughput.
  3. In-house Human Intelligence Experts acting as project managers to take care of task workflows, QA, and worker incentives.

Predictive Machine Learning:

  1. Algorithms continuously learn to improve worker proficient selection and cost optimisations.
  2. Proprietary QA algorithms.

How does Playment work?

With nothing more than data and guidelines, Playment can go out and grab trained crowd labor for any particular type of tasks. HI experts build custom instructions and manage end-to-end training models for crowd qualification. The shared data then passes through a customized workflow:Artificial Intelligence step –> To measure the confidence levelIf the confidence level is lesser than the threshold, it’s sent to the trained labor pool. There is a predictive model continuously monitoring accuracy optimization and re-runs the task(s) until the quality bar is met.Here's the example of Playment’s QA system for autonomous driving annotations,[caption id="attachment_59545" align="aligncenter" width="1693"]

Human-in-the-Loop QA Workflow Diagram for autonomous driving annotation

(open image in new tab for a better view)[/caption]

How does Traditional Crowdsourcing work?

With traditional crowdsourcing crowdsourcing like Mechanical Turk, you manually define instructions and quality metrics. Obviously, this breaks down when you are talking about multiple classes with hundreds or thousands of images, text or any kind of data. With the Playment approach, nothing is done by you, we do end-to-end project management from setting up the process (workflow) till sharing training data. We use our proprietary QA mechanism and complex machine learning models available. Therefore, you can have confidence that your data is proven to be accurate. Know more about the various image annotation services provided by Playement here.To put it simply, Playment is fully managed Human Intelligence platform for enterprise needs where traditional crowdsourcing fails.If you're interested to learn more, request a demo to talk about how we can help your business process at scale.Must read: Have you chosen the right data labeling partner for your perception model?