Back to results
US 12,748,811 B2Granted
Machine learning driven adaptive testing
AssigneeFusemachines, Inc.
InventorsShashanka Prajapati, Kiran Maiya Prajapati, Utsav Ratna Tuladhar, Kaushal Raj Mishra, Rakesh Kumar Katuwal, Sameer Maskey
Life of this patent
Estimating the term…
Abstract
The present disclosure provides a system for predicting the difficulty of an exam question in which a training module uses training data to create a machine learning model to predict the difficulty of a question, and the difficulty module collects inputted questions and answers from a teacher that are inputted through an education network into the machine learning model, and the difficulty of the question is displayed.
Claims8
A method for machine learning driven adaptive testing, the method comprising:
- receiving online testing data associated with an online course from a teacher device, wherein the online testing data includes a plurality of questions and one or more correct answers corresponding to the plurality of questions;
- providing the received online testing data to a difficulty scoring machine-learning model, wherein the difficulty scoring machine-learning model incorporates a knowledge graph that includes graph-structured data correlating words of questions to an associated difficulty level derived from historical probability-weighted online testing data;
- analyzing the plurality of questions using the difficulty scoring machine-learning model to predict a difficulty level of the plurality of questions in the online testing data for students of the online course, wherein the prediction is based at least in part on a difficulty level corresponding to words of the plurality of questions as derived from the knowledge graph and historical probability-weighted online testing data including previously analyzed questions having one or more same words as the plurality of questions in the received online testing data;
- training the difficulty scoring machine-learning model by updating the historical probability-weighted online testing data for the graph-structured data correlating words of questions to the associated difficulty level relative a percentage of correct respective answers from students associated with the online course to the predicted difficulty level;
- and
- automatically generating a future test including at least a subset of the plurality of questions, wherein the future test is responsive to a requested difficulty level corresponding to a new group of students associated with the online course and adaptively selected given a current difficulty level of previously analyzed historical probability-weighted online testing data.
3 more elements
A non-transitory computer-readable storage medium comprising instructions executable by a processor to perform a method for machine learning driven adaptive testing, the method comprising:
- receiving online testing data associated with an online course sent over a communication network from a teacher device, wherein the online testing data includes a plurality of questions and one or more correct answers corresponding to the plurality of questions;
- providing the received online testing data to a difficulty scoring machine-learning model, wherein the difficulty scoring machine-learning model incorporates a knowledge graph that includes graph-structured data correlating words of questions to an associated difficulty level derived from historical probability-weighted online testing data;
- analyzing the plurality of questions using the difficulty scoring machine-learning model to predict a difficulty level of the plurality of questions in the online testing data for students of the online course, wherein the prediction is based at least in part on a difficulty level corresponding to words of the plurality of questions as derived from the knowledge graph and historical probability-weighted online testing data including previously analyzed questions having one or more same words as the plurality of questions in the received online testing data;
- training the difficulty scoring machine-learning model by updating the historical probability-weighted online testing data for the graph-structured data correlating words of questions to the associated difficulty level relative a percentage of correct respective answers from students associated with the online course to the predicted difficulty level;
- and
- automatically generating a future test including at least a subset of the plurality of questions, wherein the future test is responsive to a requested difficulty level corresponding to a new group of students associated with the online course and adaptively selected given a current difficulty level of previously analyzed historical probability-weighted online testing data.
3 more elements
Dependencies are inferred from the claim text.