Real-time Automated Answer Scoring

Artificial Intelligence
NLP
ML4Education
Realtime Systems

A system for automated evaluation of descriptive answers, which are traditionally the domain of human assessment. I used Glove and Word2Vec word embedding models coupled with a custom DNN architecture to generate the answers' scores. This model was tested on real-world data and improved the standing accuracy from 81% to 97.2%. This project was deployed in my undergraduate university to track a student's progress and scoring trends for an answer. This project was presented at The 18th IEEE International Conference for Advanced Learning Technologies.



Technologies Used: Python, Tensorflow, Flask Web Server, NLTK.

My Role:
  • Designed the custom DNN architectures, with neural network optimization and hyperparameter search.
  • Built the web application to be a real-time system that captures ”snapshots” of the student's progress that was deployed in my university.
  • Incorporate feedback from the university's examination department to improve the real-time system.

Real-time Automated Answer Scoring

Project Gallery

sally lightfoot crab
Real-time Processing of Descriptive Answers
fighting nazca boobies
Comparison of word embeddings and models
fighting nazca boobies
Errors across 20,000 descriptive answers