Understanding graph reasoning
This section is devoted to a discussion of how to solve graph data tasks. In this section, we will discuss some of the approaches used to solve tasks on knowledge graphs: KG embeddings, GNNs, and LLMs. KG embeddings and GNNs would require at least one chapter each; hence, these topics are outside the scope of the book, but we believe that an introduction to them would be beneficial to a practitioner. In fact, both embedding and GNNs can be used synergistically with LLMs and agents.
There are many tasks in which a model is required to understand the structure to solve, and these are collectively called graph structure understanding tasks. Many of these tasks are solved using algorithms or models designed specifically to learn these tasks. Today, a new paradigm is being developed in which we try to use LLMs to solve these tasks; we will discuss this in depth at the end of this section. Examples of tasks might be degree calculation (how many neighbors...
