Suppose that you know that the nodes of a social network can be classified as follows: computer science majors, math majors, and engineering majors. You only have one labeled node for each of the three classes $--$ but you also know that students tend to make social connections with students of the same major. Which of the following approaches do you expect would perform best in this classification problem?

Compute adjacency$-$based node embeddings, and classify each node based on the shortest distance between its embedding vector and the embedding vectors of the three labeled nodes.

Compute random$-$walk node embeddings $($with a length that is more than $1$ but much less than the network diameter$),$ and classify each node based on the shortest distance between its embedding vector and the embedding vectors of the three labeled nodes.

Classify each node based on the minimum shortest$-$path length between that node and the three labeled nodes.

Use a supervised learning approach to train a deep graph neural network using the three labeled nodes.