Conference

The Sequence: Crowdsourcing natural language data

In this tutorial, leading researchers and engineers from Toloka share their unique industry experience in achieving efficient natural language annotation with crowdsourcing.

Image
Image
Image
+1
Image

Overview

In this tutorial designed specifically for the readers of The Sequence, leading researchers and engineers from Toloka will share their unique industry experience in achieving efficient natural language annotation with crowdsourcing. We will introduce data labeling via public crowdsourcing marketplaces and present the key components of efficient label collection. Then, in the practice session, participants will choose one real language resource production task, experiment with selecting settings for the labeling process, and launch their label collection project on Toloka, one of the world’s largest crowdsourcing marketplaces. During the tutorial session, all projects will be run on the real Toloka crowd. We will also present useful quality control techniques and give the attendees an opportunity to discuss their own annotation ideas.

Topics

  • Reasons for collecting and labeling data via crowdsourcing for SDC: pros & cons
  • Key components of crowdsourcing for efficient data labeling
  • Decomposition approach
  • Performers selection and training
  • Hands-on practice session: audio transcription
  • Advanced crowdsourcing techniques: aggregation, incremental relabeling & pricing
  • Example of how to aggregate crowdsourced texts using the Crowd-Kit Python library

Speakers

Image
Dmitry Ustalov
TolokaAnalyst / Software Developer
Image
Daria Baidakova
TolokaDirector of Educational Programs
Image
Natalie Fedorova
TolokaEducational Project Manager
Image
Nikita Pavlichenko
TolokaAnalyst / Software developer

Organizers

Image
Natalie Fedorova
TolokaEducational Project Manager
Image
Daria Baidakova
TolokaDirector of Educational Programs

Schedule

16:00 – 16:15

Part 0: Introduction
— The concept of data labeling via crowdsourcing
— Crowdsourcing task examples
— Crowdsourcing platforms
— Yandex crowdsourcing experience

10:15 – 10:45

Part I: Key Components for Efficient Data Collection
— The concept of data labeling via crowdsourcing
— Crowdsourcing task examples
— Crowdsourcing platforms
— Yandex crowdsourcing experience

16:15 – 16:45

Part I: Key Components for Efficient Data Collection
— Decomposition for effective pipeline
— Task instruction & interface: best practices
— Quality control techniques

16:45 – 17:45

Part II: Practice part I
— Dataset and required labels
— Discussion: how to collect labels?
— Data labeling pipeline for implementation
— You
» create
» configure
» run data labeling projects on real performers in real-time

17:45 – 18:30

Break

18:30 – 19:00

Part III: Advanced techniques
— Incremental relabeling
— Dynamic pricing

19:00 – 19:30

Break

19:30 – 19:45

Part IV: Practice part II
— Finishing up label collection
— Results aggregation

19:45 – 20:15

Part V: Conclusion
— Results of your projects
— Ideas for further work and research
— References to literature and other tutorials
— Q&A

Text Aggregation Example

We will share an example of how to aggregate crowdsourced texts using the Crowd-Kit library for Python.

(

Don't miss out

Be the first to hear about our workshops, 
tutorials, and webinars.
Fractal