Semantic Content Graph Guild

One of many goals of AI/ML-driven content goes beyond simple recommendation of individual topics. In many cases, it is perfectly appropriate to retrieve and deliver a single topic for a single purpose – individual concept topics to inform users, single-task topics to help users accomplish self-contain tasks, chunks of individual topics to provide specific answers. However, software and machines are often variable, multi-function solutions and devices.

Where we often fail is providing personalized and targeted content collections organized to address complex multi-task scenarios. The industry has focused far too much on task-oriented content to the detriment of serving multi-task scenarios. We do this as best we can in static, prescriptive organization of topics, such as with maps, submaps, relationship tables, and more, but that isn’t enough to deal with the variability of what the individual users can do with a complex product or service.

When thinking about how we might apply graph technology and machine learning, I contemplated how we might enable machines to not only find and retrieve topics and microcontent, but automatically organize them to address specific user scenarios. That led me to think about models and patterns as potential resources to accomplish such a feat. Consider the following patterns and models that might be useful to us, if not essential:

[*]The inherent prescriptive organization contained in prescriptive document object models, such as author-defined hierarchies.
[*]Other prescriptive organizational mechanisms (for example, relationship tables and links).
[*]External organization resources (macro and micro journey maps)
[*]The content use model (a mapping of topic instances (titles) to content types (both DITA type and use type) and use domains
[*]Domain-specific graph (e.g. in my company's case, a tax-compliance ontology and graph/
[*]Post Pathing: Paths customers have taken through a product.
[*]Real-time pathing: Paths customers are taking through a product.
[*]Other personalized intelligence about the user

This raises a number of questions:

• How can we leverage these assets in conjunction with a graph of our content to automatically find, retrieve, and organize that content?
• Can we use that intelligence as initial ML training as well as core patterns to initially use.
• How do we use these assets – can some of them (like multi-task scenario journeys) be encoded in a machine-consumable format and become additional graphs themselves to triangulate with the graph(s) of the content corpus?
• Can we mine one graph with another? If so, How does one do that? Or do we merge graphs?

Regards,
Michael

Yes! These are exactly the types of issues I’ve been thinking about. It’s been a while since I took SQL classes, but I keep thinking of joins—pull content (data) from different sources (tables) using a common taxonomic term shared by all sources (primary key). If you can determine what the common taxonomy (products, content types, etc.) is that all your different sources is, you can do a really solid job of defining taxonomy. Then use that as the primary key to pull data from other sources that may or may not be as well defined as the core taxonomy. Those could rely more on AI to infer relationships. The trick seems to be identifying the taxonomies that need to be rock solid (the primary keys) vs ones that can be less structured.