Measuring A Student’s Ability To Score By Tracking Knowledge Levels And Behavioural Patterns

“Everybody is a genius. But if you judge a fish by its ability to climb a tree, it will live its whole life believing that it is stupid.” — Einstein 

This quote sums up the woes of students across the world. Educational systems and recruiters consider only examinations as proof of excellence. Despite being the best or worse all year long, what a student does in an exam is the bottom line. Adding to the woes, a huge gap in demand and supply of quality teachers has resulted in a poor student-teacher ratio. Teachers cannot effectively provide personalised, individual attention to every student. To overcome all these problems, we need to provide students with what they need, when they need, and at their own pace to reach their full potential.

To achieve this, Embibe measures students’ current ability to perform in the exam using scientific parameters known as ‘Embibe Score Quotient’. Embibe, as a whole, believes a student’s ability to score depends on academic proficiency and behaviour across multiple fine-grained latent attributes. It allows us to pinpoint the lack of exact elements responsible for low test scores.

The core components of the ‘Embibe Score Quotient’ that allow our system to provide spot recommendations are:

1. Academic Quotient: It denotes students’ academic capability. At Embibe, the academic capability of a student is calculated with the help of ‘Concept Mastery’. ‘Concept Mastery’ measures a student’s mastery of a concept based on various activities performed by the student, such as watching videos, practicing questions, taking tests and reviewing test feedback.
2. Behaviour Quotient: It denotes a student’s behavioural traits concerning test-taking. It comprises sub-categories, namely Intent Quotient and Test Taking Skills Quotient.
   1. Intent Quotient denotes a student’s scholastic attitude or intent irrespective of his knowledge. Some non-exhaustive features considered in this category are the number of wasted attempts, subject swaps, time spent not attempting any question, and other similar features.
   2. Test Taking Skills Quotient denotes a student’s test-taking ability. To give a better idea, some of the features considered in this category are the number of questions looked at and the number of questions marked for review, among other similar features.
3. Effort Quotient: It denotes the measure of effort a student puts in while learning. Effort Quotient allows us to grasp the student’s potential to score in exams. Some of the directly interpretable features under this category are the number of practice sessions, total time spent in practice, and other similar activities.

Machine Learning Model for Spot Recommendation

A Machine Learning Model is an approximation of reality. An interpretation can be considered a medium to assert that reality either in quantitative or natural language. It is urged to use interpretable models where possible. There is a surge in providing interpretations to even seemingly opaque deep learning models. Keeping interpretability in mind, ‘Embibe Score Quotient’, built with rich and highly interpretable features, provides spot recommendations to students that aims to improve their weaker concepts and provide behavioural feedback in the most granular and action-oriented manner.

For example, for a student with a high Academic Quotient, high Intent Quotient but low Test Taking Quotient, the system would suggest, “You spend too much time on questions where you are unsure of the subject material. Reduce that time to allow you more time overall on your test.” Different suggestions are provided for different scenarios.

In conclusion, the ‘Embibe Score Quotient’, which is the projection of students’ ability into Academic, Behavioral and Effort dimensions, allows Embibe’s AI Engine to mitigate the low teacher-student ratio by providing personalised, individual attention to the student. Furthermore, with ‘Embibe Score Quotient’ at the core of the AI Engine, we also enhance the system’s human interpretation, bringing it closer to Explainable AI.

References:

[1] Keyur Faldu, Aditi Avasthi, and Achint Thomas. Adaptive Learning Machine for Score Improvement and Parts Thereof, US Patent No. 10854099 B2.

[2] C. Rudin. Stop Explaining Black-Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. arXiv e-prints, 11 2018.