Semantic Network

Definition

A Semantic Network is a qualitative graph that represents knowledge as a set of entities (concepts, objects, ideas) connected by meaningful, typed relationships. In the context of Hellixia and BayesiaLab, it is implemented as a special kind of Bayesian Network where nodes are continuous variables with associated vector embeddings, and arcs are semantic correlations learned from the similarity of those embeddings.

Characteristics

• Nodes represent entities (for example, a list of client organizations like “L’Oréal” or “NASA”, or conceptual dimensions of a domain like “Human Resources”, “Maintenance Status”, etc.).
• Edges are typed semantic relations – they can be things like is-a, part-of, causes, or simply a strong correlation in meaning.
• It is qualitative, not probabilistic – you do not use a Semantic Network for probability queries or inference. Its value is interpretive: it gives you a navigable map of a field of knowledge, showing which concepts cluster together and how they relate.

Purpose and Use

The primary goal of a Semantic Network is in-depth domain understanding. By arranging entities and their strongest semantic correlations in a single visual map, an analyst can quickly grasp the conceptual landscape of a field. For example:

• A Semantic Network of Clients reveals which organizations are conceptually related (e.g., companies vs. universities vs. agencies).
• A Semantic Network of Elicited Dimensions helps a team see all the factors relevant to a modeling problem and select which variables to include in a later Bayesian network.

Because the network is qualitative, you can treat it as a stepping stone to more formal models. After exploring, you may refine it into a Knowledge Graph (with explicitly typed relations), a Causal Semantic Diagram (with causal arcs), or eventually a Causal Bayesian Network (quantified for inference).

Analogy

Think of a Semantic Network as a mind map on steroids. A mind map draws a central idea and branches out to related concepts, but the connections are drawn by hand and can be subjective. A Semantic Network does the same thing automatically: it reads your list of concepts, measures how close each pair is in meaning (via embeddings), and then draws only the strongest links. The result is an objective, data-driven structure that reveals the hidden structure of your knowledge.