Deriving Character Logic from Storyline as Codified Decision Trees
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908767338102784 |
|---|---|
| author | Peng, Letian Zhou, Kun Yun, Longfei Hou, Yupeng Shang, Jingbo |
| author_facet | Peng, Letian Zhou, Kun Yun, Longfei Hou, Yupeng Shang, Jingbo |
| contents | Role-playing (RP) agents rely on behavioral profiles to act consistently across diverse narrative contexts, yet existing profiles are largely unstructured, non-executable, and weakly validated, leading to brittle agent behavior. We propose Codified Decision Trees (CDT), a data-driven framework that induces an executable and interpretable decision structure from large-scale narrative data. CDT represents behavioral profiles as a tree of conditional rules, where internal nodes correspond to validated scene conditions and leaves encode grounded behavioral statements, enabling deterministic retrieval of context-appropriate rules at execution time. The tree is learned by iteratively inducing candidate scene-action rules, validating them against data, and refining them through hierarchical specialization, yielding profiles that support transparent inspection and principled updates. Across multiple benchmarks, CDT substantially outperforms human-written profiles and prior profile induction methods on $85$ characters across $16$ artifacts, indicating that codified and validated behavioral representations lead to more reliable agent grounding. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_10080 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Deriving Character Logic from Storyline as Codified Decision Trees Peng, Letian Zhou, Kun Yun, Longfei Hou, Yupeng Shang, Jingbo Computation and Language Role-playing (RP) agents rely on behavioral profiles to act consistently across diverse narrative contexts, yet existing profiles are largely unstructured, non-executable, and weakly validated, leading to brittle agent behavior. We propose Codified Decision Trees (CDT), a data-driven framework that induces an executable and interpretable decision structure from large-scale narrative data. CDT represents behavioral profiles as a tree of conditional rules, where internal nodes correspond to validated scene conditions and leaves encode grounded behavioral statements, enabling deterministic retrieval of context-appropriate rules at execution time. The tree is learned by iteratively inducing candidate scene-action rules, validating them against data, and refining them through hierarchical specialization, yielding profiles that support transparent inspection and principled updates. Across multiple benchmarks, CDT substantially outperforms human-written profiles and prior profile induction methods on $85$ characters across $16$ artifacts, indicating that codified and validated behavioral representations lead to more reliable agent grounding. |
| title | Deriving Character Logic from Storyline as Codified Decision Trees |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2601.10080 |