Skip to content

Deep Learning on Graph between Educational Contents

Notifications You must be signed in to change notification settings

SharoneDayan/EduDeepGraph

Repository files navigation

EduDeepGraph

Graphs in Machine Learning (MVA) project about using Deep Learning on Graph between Educational Contents.
The aim of this project is to predict a student's answer to an educational question using a knowledge base.
With this base, we are building a similarity graph and graph convolutions for the predictions.

Dataset : lelivrescolaire dataset (questions/answers for each student)
Advisor : Julien Seznec (lelivrescolaire).

Building the graph

Jupyter notebook file
We have m(~25k) students and n(~15k) questions.
Adjacency matrix (W) / Graph :

  • nodes = questions
  • edges = correlations between questions (difficulty/success/spentTime mean over all students) using L1-norm

W is a n x n matrix.

Graph Convolutional Network (GCN)

Jupyter notebook file
We are using T. Kipf's paper and implementation.
Train set :

  • Adjacency matrix (W, n x n matrix)
  • History matrix : students' answers to the questions (H, n x m matrix, sparse matrix)

About

Deep Learning on Graph between Educational Contents

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published