从故事线推导出角色逻辑作为编码决策树
Deriving Character Logic from Storyline as Codified Decision Trees
AI总结:
本文提出CDT框架,通过大规模叙述数据诱导可执行且可解释的决策树结构,提升角色扮演代理行为的可靠性与可维护性。
摘要:
角色扮演(RP)代理依赖于行为档案在不同的叙述情境中保持一致的行为,但现有的档案大多是无结构的、不可执行的且验证不足,导致代理行为脆弱。我们提出了编码决策树(CDT),一种数据驱动的框架,从大规模叙述数据中诱导出可执行且可解释的决策结构。CDT将行为档案表示为条件规则的树状结构,其中内部节点对应经过验证的场景条件,叶子节点编码为基于现实的行为陈述,使在执行时能够确定性地检索上下文合适规则。该树通过迭代诱导候选场景-动作规则、验证它们并与数据对比,并通过层次化专门化进行细化,产生支持透明检查和原则性更新的档案。在多个基准测试中,CDT在16个物品上85个角色上显著优于人工编写档案和先前档案诱导方法,表明经过编码和验证的行为表示能产生更可靠的代理基础。
英文摘要:
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.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。