Skip to Content
BayesiaLabBayesiaLab User GuideMenusHellixiaDocument AnalysisSemantic Network Generator

Semantic Network Generator

Overview

The Semantic Network Generator learns a Semantic Network from a Knowledge File by extracting latent dimensions and concept embeddings, revealing how the document’s ideas cluster by meaning. Provide the file and selected keywords, and it returns the dimensions, embeddings, and a learned Semantic Network.

Use it for semantic proximity among document-derived concepts and theme discovery in source material.

Workflow

Node Generation

Produce the set of nodes that represent the domain’s entities. This can be done in three ways:

Manually, by typing the nodes by hand.
By loading a data set, so that its entries become nodes.
By using the Dimension Elicitor, which draws on the knowledge embedded in a Large Language Model to automatically propose and create nodes for the domain.
Embedding Generation

Run the Embedding Generator (via Hellixia > Embedding Generator) to build a continuous high-dimensional vector for every node from its Node Name, Long Name, and/or Comments. These embeddings capture the semantics of each node, making them comparable so that the subsequent learning step can find semantic correlations. The richer a node’s description, the more relevant its embedding.

Learning the Semantic Network

Learn the structure by running Unsupervised Structural Learning with the Maximum Spanning Tree algorithm (via Learning | Unsupervised Structural Learning | Maximum Spanning Tree). This connects the semantic nodes by edges that represent the strongest semantic similarities, producing a network that clusters semantically similar entities together.

The workflow is demonstrated in the Semantic Network of Clients example, where a large list of client organizations is loaded, embedded, and then structured with the Maximum Weight Spanning Tree, resulting in clusters such as the Consumer Goods Client Cluster (grouping L’Oreal, Procter & Gamble, Unilever, Nestle, Danone, Mars Inc., Mondelez, Shiseido, Coca-Cola, etc.).

To make the resulting network more robust, you can enhance node documentation before step 2 using tools such as the Keyword-Based Node Comment Generator or the Definition-Based Comment Generator (accessed via Comments > Definition-Based Comment Generator or Properties > Comment > Define Node Concept). Better documentation feeds the Embedding Generator and yields more faithful semantic correlations.

Examples

Propositional Causal Bayesian Network GeneratorDoc-to-Node Generator