Guide for Graph Mining Learning

Your first baby step to learn Deep Learning for Graph Network

Imagine Facebook: How do you get connected within layers of friends?

Imagine Recommendation System: How do you know a person’s preference is closely related to its clusters?

Welcome to Graph Mining

Graph classification generates graphs among a vast amount of connected data (e.g: Social, Biological, and Payment) and uses the graphs to identify labels (supervised) or clusters (unsupervised).

If this sounds tough to you, you can look no further than your brain because it is the master at inferencing connections among graphs quickly.

For example, your brain knows how to get from point A to point B without even thinking. Imagine your last grocery trip after being on shopping duty for a long time. You can skim through the catalogues and make multiple beelines of different categories of aisles to compare products while bringing your children to shop.

All these actions require consolidations to infer the best action based on series of sub actions to reach your end goal. All done in the most effortless methods as possible.

Graph mining uses features to see how a set of observations are related from a user facing similarity signal.

Graphs represent relationships (edges) between entities (nodes) which are formulated based on distance.

There are two characteristics commonly found:

  • Natural graphs which come from an external source. For example payment networks, social media, and roadways.
  • Similarity Graphs which comes from measures of similarity distance between nodes. For example a blob of metadata then shares the blob structure via graph representation.

Depending on the characteristics of the graphs, we can classify each graph as

  • Simple homogeneous with one type of node and edge
  • Complex heterogeneous with multiple types of nodes, multiple types of edges. These can be directed or undirected.

Graphs are everywhere

Graphs are multipurpose. You can build supervised models by classifying surrounding neighbors, semi-supervised models by propagating existing labels to missing labels in the neighborhood, and unsupervised models by training node-level embeddings to describe structural role of the data

There are a wide variety of tasks you can complete with graphs. For example, you can classify nodes, graphs,…

Continue reading:—-7f60cf5620c9—4