SPeCtrum: A Grounded Framework for Multidimensional Identity Representation in LLM-Based Agent

TL;DR

SPeCtrum proposes a grounded, multidimensional identity framework for LLM agents by integrating Social Identity, Personal Identity, and Personal Life Context. Automated evaluations with fictional characters show that Personal Life Context often drives accurate identity representations, rivaling full SPC configurations, while human evaluations reveal that the full SPC combination yields richer self-concept fidelity for real individuals. The work demonstrates both the potential and limits of using C alone and highlights the importance of incorporating broader data sources to capture real-world identities. These findings inform the design of more authentic, personalized AI personas and guidance for future cross-cultural and privacy-conscious deployments.

Abstract

Existing methods for simulating individual identities often oversimplify human complexity, which may lead to incomplete or flattened representations. To address this, we introduce SPeCtrum, a grounded framework for constructing authentic LLM agent personas by incorporating an individual's multidimensional self-concept. SPeCtrum integrates three core components: Social Identity (S), Personal Identity (P), and Personal Life Context (C), each contributing distinct yet interconnected aspects of identity. To evaluate SPeCtrum's effectiveness in identity representation, we conducted automated and human evaluations. Automated evaluations using popular drama characters showed that Personal Life Context (C)-derived from short essays on preferences and daily routines-modeled characters' identities more effectively than Social Identity (S) and Personal Identity (P) alone and performed comparably to the full SPC combination. In contrast, human evaluations involving real-world individuals found that the full SPC combination provided a more comprehensive self-concept representation than C alone. Our findings suggest that while C alone may suffice for basic identity simulation, integrating S, P, and C enhances the authenticity and accuracy of real-world identity representation. Overall, SPeCtrum offers a structured approach for simulating individuals in LLM agents, enabling more personalized human-AI interactions and improving the realism of simulation-based behavioral studies.

Figures (8)

• Figure 1: Overview of the SPeCtrum Framework for Multidimensional Identity Representation
• Figure 2: Guess Who & TST Evaluation Procedure
• Figure 3: Character Identification Accuracy in Guess Who Evaluation across LLMs and Conditions
• Figure 4: Accuracy of TST Statements across LLMs and Conditions
• Figure 5: Perceived Similarity Ratings across Conditions and Essay Topics