食谱也有人格吗?属性化程序图中的创作者风格表征与生成
文章背景与核心概要
大型语言模型(LLM)往往倾向于生成泛化、同质化的逻辑,难以捕捉人类创作者独特且具特异性的执行风格。本文已被第29届国际发现科学会议(DS 2026)录用,研究了如何从非结构化数据中计算发现和生成程序化人格(Procedural Personas)。作者引入了一个名为 ViralRecipesTrans 的全新数据集,该数据集将从烹饪视频文本中提取的执行流图精确映射到特定的创作者。
通过将程序化文体学(Procedural Stylometry)建模为图学习和过程发现任务,本研究揭示了一个核心二分法:传统的词汇分类器容易受语义泄露(Semantic Leakage)困扰,而离散拓扑指标则能成功捕捉创作者刻板的物理工作流。为了弥合这一差距,论文提出了一种由马尔可夫先验驱动的结构化两阶段模型,以及一个将全局语义推理与局部拓扑印记动态结合的集成框架。
论文元数据 (Paper Metadata)
- arXiv 标识符: arXiv:2608.24369 [cs.AI]
- 标题: Do Recipes Have Personas? Characterizing and Generating Creator Style in Attributed Procedural Graphs
- 作者: Lei Jiang
- 主要学科: 人工智能 (
cs.AI) - 提交日期: 2026年8月25日
- 发表会议: 第29届国际发现科学会议 (DS 2026) 录用(15页,2张图表)
- arXiv Identifier: arXiv:2608.24369 [cs.AI]
- Title: Do Recipes Have Personas? Characterizing and Generating Creator Style in Attributed Procedural Graphs
- Author: Lei Jiang
- Primary Subject: Artificial Intelligence (
cs.AI)- Submission Date: August 25, 2026
- Venue: Accepted at the 29th International Conference on Discovery Science (DS 2026) (15 pages, 2 figures)
摘要与核心贡献 (Abstract & Key Contributions)
摘要: 尽管大语言模型(LLM)拥有丰富的零样本程序化知识,但它们倾向于生成同质化的逻辑,这往往掩盖了人类创作者独特且具特异性的执行过程。在本文中,我们研究了从非结构化数据中计算发现程序化人格的方法。为此,我们引入了 ViralRecipesTrans,这是一个新的数据集,包含从热门烹饪视频文本中提取的程序化对齐执行流图,并显式映射到特定创作者。我们将程序化文体学表述为图学习和过程发现任务,揭示了一个基本的二分法:传统的词汇分类器通过语义泄露导致过拟合,而离散拓扑指标则能成功捕捉创作者工作流的严格物理约束。在此表征的基础上,我们将框架扩展到一个新的生成任务——预测未见菜肴中创作者的确切结构执行图。我们揭示了风格生成中全局宏观规划与局部结构执行之间的根本二分法。我们的结果表明,少样本 LLM 在语义分配方面占主导地位,但存在持续的宏观规划缺陷,而我们的结构化两阶段模型通过严格的马尔可夫先验实现了卓越的拓扑控制。总之,程序化生成的集成方法结合了双方的优势,动态合成全局语义推理与局部拓扑印记,以实现个性化工作流的自动化发现与生成。
Abstract: While large language models (LLMs) possess vast zero-shot procedural knowledge, their tendency to produce homogenized logic often obscures the unique, idiosyncratic execution processes of individual human creators. In this paper, we investigate the computational discovery of procedural personas from unstructured data. To achieve this, we introduce ViralRecipesTrans, a new dataset of procedurally aligned execution flow graphs extracted from popular culinary video transcripts and explicitly mapped to specific creators. We formulate procedural stylometry as a graph learning and process discovery task, revealing a fundamental duality: while traditional lexical classifiers overfit via semantic leakage, discrete topological metrics successfully capture the rigid physical constraints of a creator's workflow. Building upon this characterization, we extend our framework into a novel generative task--predicting a creator's exact structural execution graph for unseen dishes. We expose a fundamental dichotomy in style generation between global macro-planning and local structural execution. Our results demonstrate that few-shot LLMs dominate semantic assignment but suffer from persistent macro-planning deficits, whereas our structured two-stage model achieves superior topological control via rigid Markovian priors. Together, an ensemble approach to procedural generation combines the strengths from both sides, dynamically synthesizing global semantic reasoning with localized topological footprints to automate the discovery and generation of personalized workflows.
核心发现与方法论 (Key Findings & Methodology)
-
ViralRecipesTrans 数据集: 一个精选的执行流图集合,提取自烹饪视频文本,并显式映射到各个创作者。
- The ViralRecipesTrans Dataset: A curated collection of execution flow graphs extracted from culinary video transcripts, explicitly mapped to individual creators.
-
程序化文体学与二分法: 证明了传统的词汇分类器通过语义泄露产生过拟合,而离散的拓扑指标则能有效捕捉物理工作流的约束。
- Procedural Stylometry & Duality: Demonstrates that traditional lexical classifiers overfit via semantic leakage, whereas discrete topological metrics effectively capture physical workflow constraints.
-
宏观规划 vs. 局部执行: 突出了风格生成中的二分法:
- LLM: 擅长语义分配和局部内容生成,但在宏观规划方面存在困难。
-
两阶段结构化模型: 利用严格的马尔可夫先验来实现卓越的拓扑和结构控制。
- Macro-Planning vs. Local Execution: Highlights the dichotomy in style generation:
- LLMs: Excel at semantic assignment and localized content generation, but struggle with macro-planning.
- Two-Stage Structured Model: Leverages rigid Markovian priors for superior topological and structural control.
-
集成程序化生成: 将 LLM 驱动的全局语义推理与局部拓扑印记相结合,为未见过的菜肴生成个性化的工作流。
- Ensemble Procedural Generation: Combines LLM-driven global semantic reasoning with localized topological footprints to yield personalized workflows for unseen dishes.