SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation

arXiv cs.CL / 4/13/2026

📰 NewsDeveloper Stack & InfrastructureIdeas & Deep AnalysisModels & Research

共有:

Key Points

SPASM is introduced as a stability-first framework for generating multi-turn synthetic dialogues where LLM agents must maintain consistent personas, roles, and goals over long horizons.
The approach modularizes persona creation (schema sampling and validation), client–responder dialogue generation, and termination detection to produce coherent, well-scoped conversations.
To prevent long-horizon identity failures like persona drift, role confusion, and “echoing,” SPASM proposes Egocentric Context Projection (ECP), which stores dialogue history in a perspective-agnostic form and deterministically projects it into each agent’s viewpoint.
Experiments across multiple LLM backbones and nine client–responder pairings generate a dataset of 4,500 personas and 45,000 conversations, with ablations indicating ECP reduces persona drift and eliminates echoing under human validation.
The authors release code for SPASM on GitHub, enabling researchers and developers to apply the framework without changing model weights.

Abstract

Large language models are increasingly deployed in multi-turn settings such as tutoring, support, and counseling, where reliability depends on preserving consistent roles, personas, and goals across long horizons. This requirement becomes critical when LLMs are used to generate synthetic dialogues for training and evaluation, since LLM--LLM conversations can accumulate identity-related failures such as persona drift, role confusion, and "echoing", where one agent gradually mirrors its partner. We introduce SPASM (Stable Persona-driven Agent Simulation for Multi-turn dialogue generation), a modular, stability-first framework that decomposes simulation into (i) persona creation via schema sampling, plausibility validation, and natural-language persona crafting, (ii) Client--Responder dialogue generation, and (iii) termination detection for coherent stopping. To improve long-horizon stability without changing model weights, we propose Egocentric Context Projection (ECP): dialogue history is stored in a perspective-agnostic representation and deterministically projected into each agent's egocentric view before generation. Across three LLM backbones (GPT-4o-mini, DeepSeek-V3.2, Qwen-Plus) and nine Client--Responder pairings, we construct a dataset of 4,500 personas and 45,000 conversations (500 personas X 10 conversations per pairing). Ablations show ECP substantially reduces persona drift and, under human validation, eliminates echoing; embedding analyses recover persona structure and reveal strong responder-driven interaction geometry. Our code is available at https://github.com/lhannnn/SPASM.