Dialogue-oriented Interpretable Personality Recognition via Evidence-Guided Bidirectional Iterative Optimization

ACL ARR 2026 January Submission10177 Authors

Keywords: Interpretable Personality Recognition, Generative-Discriminative Feedback Refinement, Evidence-Guided Bidirectional Iterative Optimization

Abstract: Identifying personality from dialogue can improve interpretability and adaptability for human–computer interaction and psychological assessment. Existing research focuses on modeling emotional trajectories and interaction patterns from entire dialogue, failing to predict personality from the specific evidence, which may serve as key clues for reasoning and enabling accuarate personality prediction. How to establish an adaptive iterative mutual reinforcing mechanism between evidence and personality is a key challenge. This paper proposes a Generative–Discriminative Feedback Refinement mechanism for dialogue-based personality prediction, it constructs a hierarchical dialogue graph to jointly model the speaker’s role, contextual dependencies, and heterogeneous interaction relationships for evidence utterance mining. Then, the generator simulates evidence utterances at different trait levels, and the discriminator derives a consistency-based judgement between the generated and original utterances to refine the initial LLM-based prediction. And the updated prediction is fed back to the graph model via bidirectional iterative optimization, improving interpretability and overall performance. Experimental results on public dataset demonstrate that the proposed method achieves the best performance over the state-of-the-art model.

Paper Type: Long

Research Area: Computational Social Science, Cultural Analytics, and NLP for Social Good

Research Area Keywords: NLP tools for social analysis

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: Chinese

Submission Number: 10177