08 Graph Based Machine Learning

Question 1

what are the types of graph data

Answer 1

networks and graphs

Question 2

what are the graph based ML tasks

Answer 2

1. node classification
   - predict property
2. edge prediction
   - predict missing link between nodes
3. graph classification
   - categorise different graphs
4. clustering
   - discover communities
5. graph generation
   - generate new graph structure

Answer 3

raw data -> feature engineering AND graph data -> learning algorithm -> model

Question 4

what are the node features

Answer 4

1. node degree
   - v = node
   - k(v) = edges that v has
2. node centrality
   - eg. closeness
3. clustering coefficient
   - how connected node V’s neighborhood of nodes are
4. graphlets
   - graphlet degree vector counts the number of graphlet that a node participates in

Question 5

what are the edge features

Answer 5

1. distance based
   - measure distance between nodes
2. local neighborhood overlap
   - measure the degree of neighborhood overlaps locally
3. global neighborhood overlap
   - measure the degree of neighborhood overlaps globally

Question 6

what is the kernel method

Answer 6

1. design kernels instead of feature vectors
2. measures similarity between data points

Question 7

what is graph based machine learning

Answer 7

focus on developing algorithms for analysing data that can be represented as graphs or networks. data is represented as nodes connected by edges. relies on structural information present in graphs to make predictions, perform clustering, classification…

key concepts
1. first step is to represent data as a graph. Nodes represent entities, edges represent relationships

2. node embedding aims to map each node in the graph to a low dimensional vector representation
3. graph convolutional networks (GCN) extend the concept of convolutional neural networks to graph structured data. applies convolutional operations to nodes in the graph
4. graph clustering aims to partition nodes in graph into groups of clusters based on their similarities or connections.
5. link prediction predicts missing/ future connection between nodes
6. graph neural networks are deep learning modesl designed to operate on graph structured data. combination of node embedding and graph convolutional layers
7. graph based recommender systems provide personalised recommendations by modeling user-item interactions

Question 8

why use node embedding