ARY2260 / Eval-1

An AI-based Automatic Test-Grader

Eval

This is Eval! A tool used to automate the evaluation of test answers using Cohere powered APIs and scoring the answers based on suitable metrics.

The REST API for the project is currently hosted on: https://replit.com/@AsimJawahir/CohereAPI

Motivation

Evaluation of answer sheets conducted in the traditional way is a tedious and critical task. Lot of time and energy is invested by the teachers or examiners to thoroughly check answers for every question. There is a likelihood of showing bias by the examiners based on personal or situational circumstances.

The overall process of evaluations causes delays in declaration of results and hence leads to slow progress in assessment based learning. Large volumes of answer sheets checking can also affect the mental health of teachers which could lead to poor teaching performance.

Scoring Metrics

1. Semantic Search - this is the primary scoring strategy of Eval. It is used to sematically understand the answer given and evaluate based on content rather than simply scoring based on textual similarities.
   • Cohere Embed was used to generate embeddings for 5 suggested answers for the question and the answer to be checked. Then we find the distance from the nearest neighbour out of the 5 suggestions and the answer. This distance is used to grade the answer.
2. Duplication Check - partially correct answers with duplication of text tended to get higher similarity scores compared to the ones without duplication.
   • To stop students from using this exploit to gain extra marks, a duplications checker was implemented based on Jaccard-Similarity between sentences within the answer.
3. Grammar Check - this strategy aims to check the grammar of the answer and assign a score based on the number of grammatical errors.
   • We used Cohere Generate endpoint to generate a grammatically correct version of the answer, then check for cosine similarity of the generated version with original version to check if the original version was grammatically correct.
4. Toxicity Check - this aims to detect for toxic content in the answer and penalize an answer if it is toxic.
   • We trained a custom classification model on Cohere using Social Media Toxicity Dataset by SurgeAI which gave a 98% precision on the test split.
5. Custom Checks - this allows for users to give different weights to each of the three different metrics based on how important they are for the evaluation of the answer. This allows for a more personalized evaluation of the answer.

Built With

• Cohere API
• Scikit-learn
• Django Rest Framework
• NextJS
• Tailwind CSS
• Flask

References

• Database Schema: https://dbdiagram.io/d/63879732bae3ed7c4543e0b1
• Figma: https://www.figma.com/file/9xSF3sm9VhrWL3TJjzg22Y/Untitled?node-id=0%3A1&t=kRUjtBC1c4ot2wHS-1
• Presentation: https://docs.google.com/presentation/d/1ZZz6VU6SPtiwoCv635J52lWzc1SG9zbq44IJe_BSRl4/edit#slide=id.g1c0dc6bdf95_0_29

Installation

Refer to INSTALLATION.md for installation instructions.