Dimension Elicitor
Overview
The Dimension Elicitor is a tool for automatic node creation that elicits relevant dimensions of a problem domain by drawing on the knowledge embedded in the deep neural networks of Large Language Models (LLMs). Rather than manually typing each node, you can use the Dimension Elicitor to have an LLM propose and create new nodes for your network.
Workflow
Select the nodes that the elicitation will operate on or expand from. This anchors the LLM query to your focus area.
Choose the keywords that will be used for sensing the domain. These act as probes against the LLM’s embedded knowledge, steering which dimensions and entities are elicited. For example, in the Renal Colic example, keywords are organized into groups such as Domain and Risk to extract differently themed dimensions.
Select the chat or text-completion LLM that will run the task. The choice governs the quality and style of generated dimensions, node comments, and analyses.
Define a background description and optionally provide a Knowledge File (a text file in formats such as txt, pdf, Word, Excel, RTF, HTML, XML, PowerPoint, mp3, pcm, wav, aac, opus, or flac) to frame the LLM query and improve relevance.
Specify whether the elicited information should populate the Name, Long Name, or Comment of the node. Targeting the Comment, for instance, enriches the text that the Embedding Generator later reads.
Choose the natural language for the output and control how long the generated descriptions or comments should be.
The Dimension Elicitor is one of the three node-generation options in the Semantic Network Workflow (the others being manual entry and loading a data set). After the Dimension Elicitor populates your network with LLM-proposed entities, you proceed to step 2 (generating Embeddings for each node) and step 3 (unsupervised structural learning, e.g., using the Maximum Weight Spanning Tree) to build a qualitative network for understanding a domain.