What is a Core Semantic Model and how is it Constructed?

To read the previous entry in this series from the [A] Semantics team for more context, click here.

What is a Core Semantic Model?

[A] defines the Core Semantic Model as: the comprehensive model that defines an organization’s terms and terminology, as well as their relationships, to allow for the identification of those terms across a domain and across systems. The CSM applies these definitions to content by way of the Core Metadata Record (CMR), a component of the CCM focused on the content asset’s metadata.

There are four key components of this definition:

1. Semantic definitions — The Core Semantic Model includes canonical definitions of all key terms to control the use of these terms throughout the organization. These definitions can come from an enterprise’s taxonomies and glossary definitions.
2. Relationships between terms — The Core Semantic Model uses accepted standards like the Simple Knowledge Organization System (SKOS) to model the semantic relationships between terms in a computer-readable way. These relationships can be used to extend content’s capabilities for multichannel delivery, for example.
3. Facilitates the identification of terms across a domain and across systems — The semantic layer represented by the Core Semantic Model allows a system to determine which terms mean the same thing, and deliver the correct term to the right channel.
4. Integration with the CMR in the CCM — A CSM serves an important function in conjunction with a CCM. The CCM provides the structure of the content, while the CSM provides the meaning and interpretation of that content. More formally, the CSM integrates with the CCM by linking the semantic definitions and relationships it contains as metadata defined in the CMR component of the CCM.

 

The Components of a Core Semantic Model

What’s a Core Semantic Model made of? Creating a computer-interpretable map of meaning certainly needs a logical basis for its construction. Here are the components [A] includes in a CSM: Taxonomies and SKOS.
 

Taxonomies and SKOS

A CSM begins with the organization’s existing taxonomies. These are likely already developed to aid in labeling and navigating the organization’s content, or to enhance search engine use within the organization. Once an audit of the organization’s taxonomies is complete, further taxonomies are built out as necessary to cover all content assets that will benefit from a semantic layer.

A taxonomy includes hierarchical relations between its terms. It indicates which terms are broader or narrower than other terms. For example, an organization may have a category of products kitchen cutlery. This category may be classified as narrower than the category kitchen tools, and also broader than the category kitchen knives. But to be completely functional and useful across sets, a Core Semantic Model will need to go further and incorporate the modeling relations from the widely applied standard for a simple knowledge organization system (SKOS).

Specifically, the model will need the ability to state when distinct terms mean the same thing, identify which terms are semantically related to one another, and identify salient collections of terms. Continuing with our example involving kitchen cutlery, by extending our model to include SKOS relations we can include further key information. We can model the fact that ‘silverware’ is a common alternative term for ‘kitchen cutlery.’ Since customers who buy kitchen knives are likely to be interested in buying a cutting board, we can also note that kitchen knives are related to cutting boards. Finally, we might wish to flag that kitchen knives require precaution when used, and do so by collecting kitchen knives with other products that require precaution.

Semantic Model Definitions

The next component of a CSM is a canonical controlled list of definitions for all semantically relevant terms. This is linked back to the Core Content Model, providing a way for content developers to understand the integrated relationship between the two models.

Relations between Terms

For greater functionality, a CSM will eventually expand to model more complex semantic relations. In moving beyond the relations represented in a taxonomy or SKOS model, the CSM will eventually develop into a complete ontology, that is, a machine-readable formal model of a domain of terms with a wider array of relations than in a taxonomy or SKOS model. With an ontology, we have a machine-readable way of expressing nearly any relation between terms that we wish. Returning to our kitchen knives example, in our SKOS semantic model, we could model that kitchen knives are related to cutting boards. But if we expand the relations we use, we can be more specific. We could use a more specific relation like used together: kitchen knives and cutting boards are used together. Using this more specific relationship between kitchen knives and cutting boards can then be leveraged to increase sales of kitchen knives, by targeting customers that purchase cutting boards.
 

How is a Core Semantic Model constructed?

Evaluation of Existing Taxonomies and sources of semantic truth

Many organizations already have taxonomies, tag sets, term dictionaries, SEO guidelines, terminology management systems, product information metadata, and endless other sources of semantic truth that can be derived from existing content assets. So, the first step to constructing a CSM involves audit these existing taxonomies and other semantic sources. These audits provide the initial set of terms used to begin modeling their relationships to the metadata fields in the Core Content Model. These taxonomies will comprise the initial foundation of our CSM.

Integration with the Core Content Model

After an audit has been made of the organization’s existing taxonomies, we can turn to the CCM of the organization’s content assets as a guide to building the CSM. This will ensure that the CSM is properly integrated with the CCM, the key to enabling intelligent customer experiences. Integrating the CSM with the CCM can be done with the following steps:

• Content Asset Inventory: We answer the following questions:
• What assets are modeled in the CCM?
• What assets have the most semantic importance?
• How are they used?
• How are they delivered?
• Where are they created?
• Semantic Analysis

Once an inventory of the content assets has been made, we can begin providing a semantic analysis. This is where we identify the crucial semantic information that we will ultimately want to model in our CSM, by answering the following questions:

• What terms form a natural category together?
• What sorts of relationships exist between terms?
• What terms are used interchangeably by the organization?
• Semantic Model Design

After the analysis is complete, we can begin constructing the Semantic Model itself. This includes a construction of a taxonomy of all semantically relevant terms, indicating hierarchical relations between terms. It also involves modeling which terms are narrower or broader than other terms.

This also includes other semantically relevant relations between terms, such as the SKOS relations discussed above, or the more specific relations that make up an ontology. As mentioned before, it is crucial that the CSM model abide by established industry standards. SKOS is one such standard. As the model develops into an ontology, the modelers will need to be attentive to other relevant standards such as RDF or OWL.

Finally, the model also includes a controlled list of canonical semantic definitions of all terms.

Often this work includes collaboration with various stakeholders to arrive at shared semantic constructions. Not every group needs to agree. Thesauri can map together what each group needs to call things for their purposes. But the essence of the meanings need a shared basis in agreed truth; and that shared truth is what we codify into the semantic model.

Semantic Model Demonstration

The value of a CSM needs to be made tangible through demonstration. Thus, the next step is to create paradigm use cases for the semantic model. For example, a finalized taxonomy can be used to aid in accessing or navigating content assets, or a model of alternate labels for terms can be used to ensure that content is consistently labeled.
The demonstration serves two important roles. First, it provides an opportunity for the stakeholders in the CSM to see the concrete value provided by the model. Second, it provides a crucial opportunity to quality test the CSM. It is often difficult to have insight into how well a semantic model will perform without testing it ‘in the wild.’


The demonstration implementation can then be used to inspire additional capabilities development initiatives building the success and example of early demonstration. Then, the semantics integrations start eventually taking on a life of their own, all sharing the same understanding of meaning provided through the CSM.