This preprint shows how pseudo factor analysis mean structures can be obtained to complement the factor structures from pseudo factor analysis.
Papers from multiple independent groups now confirm that pre-knowledge of scale structures is obtainable by factor analysing embedding similarities.
No article has presented a unified framework showing how item locations, invariance, and group differences can be checked. This needs a mean vector.
Building a predictive model of the means or generating LLM responses to average involves mixing modeling paradigms. I prefer everything in a unified framework.
The proposal replaces item means with scalar proxies that are projections of item embeddings onto difference vectors representing semantic intensity.
The intensity as a difference vector between low and high trait levels idea is loosely inspired by the king - man + woman = queen style research in NLP.
Figure 1. Intercept correlation plot from the pre-print.
Next section
From vision to Large Multi-Modal Models (LMMs) for visual reasoning parameter pre-knowledge
Last section
Pseudo factor analysis results
Return home
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
