Textual Dimension Elicitor
Work in Progress
Overview
The Textual Dimension Elicitor is a Hellixia tool that performs keyword-driven extraction from Text Variables (free-text columns loaded during data import). It uses the same logic as the Dimension Elicitor used in expert brainstorming, but instead of querying human experts, it queries an LLM to extract propositional dimensions from the comments.
Functionality and Output
Purpose
It turns unstructured reviews (or any free-text comments) into candidate variables for modeling. By choosing keywords (for example, PROS and CONS), the tool drives the creation of propositional nodes that represent the dimensions evoked in the comments.
Output
The Textual Dimension Elicitor creates propositional nodes, each representing a proposition or dimension evoked in the free text (e.g., a specific ‘pro’ or ‘con’ of a product). By default, these nodes have / states, derived by discretizing a continuous evocation score.
Downstream Step
The propositional nodes created by the Textual Dimension Elicitor are then quantified by Textual Imputation, which adds new continuous columns to the internal dataset reflecting how strongly each comment evokes each proposition. Once quantified, these nodes can be used by any BayesiaLab machine learning algorithm.
Relationship to the Dimension Elicitor
The Textual Dimension Elicitor uses the same logic as the Dimension Elicitor employed in expert brainstorming: a keyword-driven query produces candidate dimensions/nodes. The difference is the source (free text versus live experts), while the extraction mechanism and the resulting node creation are analogous.
Prerequisite
Before using the Textual Dimension Elicitor, you must have Text Variables in your dataset. Free-text columns load automatically as Text Variables when you import data, and crucially they create no nodes at import – the raw comments simply reside in the internal dataset. Hellixia then uses them for downstream extraction via the Textual Dimension Elicitor and Textual Imputation.