Anthill Inside 2019

The Problem

• Quizzes/Questions are required by students for self-evaluation and by teachers to test students.
• However, quiz creation currently requires manual effort and understanding of the topic which making the process difficult to adapt and scale.
• Multiple-Choice Questions have been proved to evaluate student’s understanding of concepts and can exhibit broad ranges of difficulty and therefore, test different levels of understanding.
• However, MCQs require challenging options which can be cumbersome to pick.

The Solution

• Using NLP to generate Multiple-Choice Questions from plain-text
  • Statistically determines best entities to blank
  • Generate semantically similar options based on context and corpus
  • Better than subjective questions MC questions tests
• A user-interface to generate quizzes
  • From any document
  • On any topic in any genre (Educational or Recreational)
  • From any paragraph

Outline

• Inspiration
• Problem Statement
• How it’s done
  • Brief introduction to NLP and NER (Named Entity Recognition)
  • Using Named Entity Recognition to pick questions
  • Ranking potential questions
  • Picking relevant/similar options to accompany answer in MCQ
  • Deploying the solution
• Scope and application of the solution

Speaker bio

• Machine Learning Engineer at Freshworks since 2019. Work on building AI models for Freshsales, our Sales CRM.
• Worked with a number of startups on NLP and ML problems.
• Google Summer of Code‘18 under Debian

Slides

http://vishalgupta.me/deck/blanq