concept proper graph model in category graphs

This is an excerpt from Manning's book Graph Powered Machine Learning.
A data infrastructure which copes with “velocity” has to provide fast access to small portion of data. For instance, suppose that you are going to implement a “real-time” recommendation engine for holiday houses renting platform. The learning algorithm uses the last users clicks and searches to suggest houses. In such case the engine doesn’t have to perform accesses to the entire dataset. It checks the last N number of clicks or the clicks in the last X size timeframe and make the prediction. A native graph database (described later in this chapter) maintains the list of relationships for each node. So starting from the last click - and having a proper graph model - the engine can navigate back to the last clicks just following the relationships of each node back. This is just an example of the high performant access for small portion of dataset that graphs provide to serve velocity.
This high-level diagram highlights how the entire process can be based on graphs. This approach doesn’t require complex or fancy algorithms for providing recommendations. With a proper graph model, a simple query can do the job. The data already contains enough information, and the graph structure helps to compute the score and returns the ordered list to the user with no need to prebuild any model: the description and prediction models overlap.