Link Prediction and viewing new edges in networkx

SLR · December 27, 2019, 5:15pm

I have used this tutorial for link prediction using DGL on my own dataset:

zheng-da/dgl-tutorial-full/blob/master/basic_apps/BasicTasks_mxnet.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this tutorial, we are going to demonstrate some basic tasks in graph learning. In general, many of the graph learning problems can fall into the following categories:\n",
    "\n",
    "* Node classification: assign a label to a node.\n",
    "* Link prediction: predict the existence of an edge between two nodes.\n",
    "* Graph classification: assign a label to a graph.\n",
    "\n",
    "Many real-world applications can be formulated as one of these graph problems.\n",
    "* Fraud detection in financial transactions: transactions form a graph, where users are nodes and transactions are edges. In this case, we want to detect malicious users, which is to assign binary labels to users.\n",
    "* Community detection in a social network: a social network is naturally a graph, where nodes are users and edges are interactions between users. We want to predict which community a node belongs to.\n",
    "* Recommendation: users and items form a bipartite graph. They are connected with edges when users purchase items. Given users' purchase history, we want to predict what items a user will purchase in a near future. Thus, recommendation is a link prediction problem.\n",
    "* Drug discovery: a molecule is a graph whose nodes are atoms. We want to predict the property of a molecule. In this case, we want to assign a label to a graph."
   ]
  },
  {

This file has been truncated. show original

My question is once I have done the training, how do I see the new predicted edges in Networkx? I’m unclear whether my graph is so small that no new edges were predicted or whether I need to store the predictions during training in order to see the new edges?

Thanks!

zhengda1936 · December 28, 2019, 6:10pm

in the inference, we predict if a node connects with another node. That is, the model can tell you which node is more likely to connect to a specified node. We didn’t add an actual edge in the graph. I hope this answers your question.

SLR · December 30, 2019, 4:00am

Could you help me moving forward with the prediction to all nodes in the graph. If I wanted to create a threshold for the prediction of all nodes in the graph and add edges based on that threshold how would I go about that? Is that possible?

Thanks!