One major advantage of the graph abstraction allows us [21st Century Humans] to gain quick insights about the flowing data of our systems. Mainly, knowledge graphs provide us with higher order abstractions. The abstractions provide an interface between our queries (and natural language questions) and our data.
Interesting pages:
– Königsberg (Ch 1. Page 3)
– The Property Graph Model (Ch 1. Page 4)
– Multiple Layers of Graphs (Ch 2. Page 19)
– Sophisticated Layering of Ontologies and Taxonomies (Ch 2. Page 22)
– Metadata and Data Layers (Ch 3. Page 29)
– The ‘Data Fabric’ Pattern (Ch 3. Page 31)
– Popular Use Cases for Actioning Knowledge Graphs ( Ch 3. Pages 35-36)
– A Simple Data Ingest Pipeline Pattern (Ch 3. Page 34)
– Anatomy of Decisioning Knowledge Graphs (Ch 4. Page 43)
– ML Workflows for Graphs (Ch 4. Page 48)
– Decisioning Knowledge Graph Use Cases (Ch 4. Pages 50-51)
– Explainability (Ch 5. Page 57)
– Contextual AI Example Data Model of Simplified Smart Home (Ch 5. Page 58)
– Chatbot System Diagram (Ch 5. Pages 65-66)
– Advanced Patterns (Ch 6. Page 72)
– Digital Twin Examples (Ch 6. Pages 74-75)
– More Resources (Ch.7 Page 78)
Key Takeaways:
Building abstractions correctly solves use cases:
For example, create a custom ontology for the recommendation layer of a knowledge graph.
– Shift from ‘what items exist in my graph?’ to ‘how do I find similar items to the one that is missing?’
– Shift from descriptive system to assistive system.
“For example, if our ecommerce retailer ties its product hierarchy into stock control data through another ontological layer, it has a way of offering other good choices to the user when the current item is out of stock or to recommend products that have better margins. All of this comes at the modest cost of writing down how the business works as a machine-readable ontology.” (Ch 2. Page 22)
Build a separate graph ‘layer’ that describes metadata of the primary graph data:
For example, describe the source of the data and the related authors, data curators, and context, etc.
“Importantly, data architects can implement this technique in a noninvasive manner with respect to the source systems containing customer data, by building it as a layer above those systems. A popular example of this is to build a knowledge graph of metadata that describes data residing separately in an otherwise murky data lake.” (Ch 3. Page 29)
Examine the 2nd Degree connections when attempting to understand social data:
For example, understand subgraphs of 2nd Degree connections to create a “Single Encompassing View” or a total description of an object (likely a person, but sometimes something else: location, product, event, or general object, etc.)
“It’s worth noting at this point that beyond helping with discovery and exploration, [social] relationships are highly predictive … However, it is remarkable that a researcher can make this prediction even more accurately based on our friends-of-friends behavior … [and] the behavior of our friends-of-friends, whom we may not know that well or at all, is more predictive of our behavior than information that pertains only to us. For more information on the science underlying social graphs, see Connected by James Fowler and Nicholas Christakis (Little, Brown and Company, 2009).” (Ch 4. Page 42)
Graph Queries and/or Graph Algorithms:
Depending on use case:
– Graph queries are useful in determining a subgraph with a particular shape (see: SubGraph[‘the enemy of my enemy is my friend’])
– Graph algorithms are useful for applying a metric globally across the entire graph (see: ‘PageRank algorithm’ for influential nodes)
“The most well-known graph algorithms fall into five classic categories:
• Community detection for finding clusters or likely partitions
• Centrality for determining the importance of distinct nodes in a network
• Similarity for evaluating how alike nodes are
• Heuristic link prediction for estimating the likelihood of nodes forming a relationship
• Pathfinding for evaluating optimal paths and determining route quality and availability” (Ch 4. Page 46)
Streaming data and augmented imagination:
For example: interactivity requires different levels of abstraction where algorithms run at different speeds based on technological constraints
“It’s important to understand that interactive speeds prohibit [the] use of global algorithms and ML training … on a per-request basis. Instead, [they] operate on a different cadence to real-time queries. The more expensive processing runs in the background, continuously enriching the actioning knowledge graph, while real-time queries get better results over time as the underlying knowledge graph improves.” (Ch 4. Page 51)
Simulation and Forecasting:
For example: The ideas surrounding the concept of a digital twin, where that twin is a graph (represented in, at least, business and/or game theory use cases)
“The enterprise
digital twin provides key values:
• A map of the business or a
smaller business unit (departmental) or layer (such as IT) of the
business
• Real-time understanding of the business
• A
way of exposing the model to pressures to see how it reacts
With a digital twin acting as a smart map, it is possible to create a rich virtual view of the business that closely matches reality. Our organizing principle follows accordingly: the elements in the real world and how they interrelate are captured in high fidelity as a property graph, and the constraints and rules that govern the real world become constraints and queries for that property graph.” (Ch 6. Page 70)
Thought-provoking Idea:
The reality of our technological context shows us how our advantages in the short run often signify meager advantages in the long run. The potential good news, if we are the underdog, is that we only need to be more innovative through our imagining of better systems and techniques. Once one becomes King, though, our competitive technological supremacy only lasts until an opponent makes the next better advancement. There’s a technological hyperinflation, and investing in our future appears to be the only way to stay relevant.