发表机构
University of Illinois Urbana-Champaign; U.S. Bank; Stanford University（伊利诺伊大学厄巴纳-香槟分校; 美国银行; 斯坦福大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出CARD框架，通过规划器、生成器与校准循环，结合多维度控制项生成逼真信用卡讨论线程，经多指标评估，其效果优于多种LLM模拟基线。
AI 中文摘要
在线信用卡讨论为研究消费者如何交流金融产品提供了自然场景，模拟这类讨论不仅需要生成单个评论，还需让生成的讨论线程符合真实用户的表达与互动方式。我们推出CARD（Controlled Agentic Reddit Discussions）框架，用于生成逼真的信用卡讨论线程。给定一条信用卡帖子及其匹配的真实线程，CARD采用非逐字指导，涵盖回复结构、评论功能、立场、语气及对话多样性。其中规划器组织这些控制项，生成器构建讨论内容，校准循环则更新评论群体，以缩小生成线程与真实线程分布的差异。我们在真实Reddit信用卡讨论上，使用词汇、语义、行为及结构指标对CARD进行评估，结果显示，CARD在多个大语言模型（LLM）上，相比模拟基线能更好地匹配真实信用卡讨论的分布，且在各项指标上呈现更小的效应量与分布距离，表明结构化规划与针对性修订可生成具备逼真度的模拟信用卡讨论。
英文摘要
Online credit card discussions provide a natural setting for studying how consumers communicate about financial products. Simulating these discussions requires more than just generating individual comments, the generated threads should also match how real users express themselves and interact with others. We introduce CARD, a framework for generating realistic credit card discussion threads. Given a credit card post and its matched real thread, CARD uses non-verbatim guidance on reply structure, comment function, stance, tone, and conversational variation. A planner organizes these controls, a writer generates the discussion, and a calibration loop updates comments' populations that contribute to differences between the generated and real thread distributions. We evaluate CARD on real Reddit credit card discussions using lexical, semantic, behavioral, and structural metrics. CARD matches the distributions of real credit card discussions better than simulation baselines across multiple LLMs and also demonstrates smaller effect sizes and distribution distances across metrics. These results show that structured planning and targeted revision can generate the realism of simulated credit card discussions.
