Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synonymix: Unified Group Personas for Generative Simulations

Huanxing Chen, Aditesh Kumar

Abstract

Generative agent simulations operate at two scales: individual personas for character interaction, and population models for collective behavior analysis and intervention testing. We propose a third scale: meso-level simulation - interaction with group-level representations that retain grounding in rich individual experience. To enable this, we present Synonymix, a pipeline that constructs a "unigraph" from multiple life story personas via graph-based abstraction and merging, producing a queryable collective representation that can be explored for sensemaking or sampled for synthetic persona generation. Evaluating synthetic agents on General Social Survey items, we demonstrate behavioral signal preservation beyond demographic baselines (p<0.001, r=0.59) with demonstrable privacy guarantee (max source contribution <13%). We invite discussion on interaction modalities enabled by meso-level simulations, and whether "high-fidelity" personas can ever capture the texture of lived experience.

Synonymix: Unified Group Personas for Generative Simulations

Abstract

Generative agent simulations operate at two scales: individual personas for character interaction, and population models for collective behavior analysis and intervention testing. We propose a third scale: meso-level simulation - interaction with group-level representations that retain grounding in rich individual experience. To enable this, we present Synonymix, a pipeline that constructs a "unigraph" from multiple life story personas via graph-based abstraction and merging, producing a queryable collective representation that can be explored for sensemaking or sampled for synthetic persona generation. Evaluating synthetic agents on General Social Survey items, we demonstrate behavioral signal preservation beyond demographic baselines (p<0.001, r=0.59) with demonstrable privacy guarantee (max source contribution <13%). We invite discussion on interaction modalities enabled by meso-level simulations, and whether "high-fidelity" personas can ever capture the texture of lived experience.

Paper Structure

This paper contains 29 sections, 2 figures.

Table of Contents

  1. Introduction & Background
  2. Methodology
  3. Life Story Graph Extraction
  4. Privacy-Preserving Graph Aggregation
  5. Graph Traversal Sampling
  6. Results and Analysis
  7. Data Preparation
  8. Evaluation Design
  9. Evaluating Behavioral Signal Preservation
  10. Evaluating Privacy
  11. Synonymix Persona Performance
  12. Behavior Signal Preservation
  13. Privacy Preservation
  14. Discussion
  15. Generating Synthetic, High-Fidelity Agent Banks from Grounded Data
  16. ...and 14 more sections

Figures (2)

  • Figure 1: Unigraph Traversal on 4 Personas
  • Figure 2: TOP: Box-plot comparison of aggregate enrichment (D→L) and transformation (L→F) distances across all GSS items. Diamonds indicate means; horizontal lines indicate medians. Two enrichment outliers ($0.4$, $0.5$) for nominal questions not shown for scale. BOTTOM: Per-item comparison of transformation vs. enrichment distance. Points below the diagonal indicate items where the pipeline loses less signal than narrative expansion adds.