AnvaMiba/llm-bargaining-scenarios
收藏资源简介:
LLM Bargaining Scenarios 是一个用于文本生成任务的数据集,专门针对讨价还价和谈判场景,基于大型语言模型(LLM)作为代理在部分信息下的行为研究。该数据集来源于论文《Used Car Salesbots? Honesty and Credulity of LLMs as Bargaining Agents under Partial Information》,包含商品讨价还价场景。每个场景描述一个交易商品(如产品名称和详细描述)、买家角色(描述买家的情境和动机)、卖家角色(描述卖家的情境和动机),以及买卖双方的保留价格范围。这些价格范围用于在每次试验中采样私人保留价格,确保买方的保留价格高于卖方,以模拟真实谈判环境。数据集总共有4561个场景,按价格层级(低、中、高、非常高)组织,其中低层级包含1516个场景,中层级869个,高层级886个,非常高层级1290个。实验部分使用低层级的前十个场景。数据以JSONL格式提供,每个场景对象包括产品名称、产品描述、买家角色、卖家角色、卖家保留价格范围和买家保留价格范围等字段。
LLM Bargaining Scenarios is a dataset for text-generation tasks, focusing on bargaining and negotiation scenarios, based on the study of large language models (LLMs) as agents under partial information. It is used in the paper Used Car Salesbots? Honesty and Credulity of LLMs as Bargaining Agents under Partial Information and contains commodity-bargaining scenarios. Each scenario describes a traded item (e.g., product name and description), a buyer persona (describing the buyers situation and incentives), a seller persona (describing the sellers situation and incentives), and reservation-price ranges for both parties. These ranges are used to sample private reservation prices per trial, ensuring the buyers reservation price is higher than the sellers to simulate real negotiation environments. The dataset totals 4561 scenarios organized by price tiers (low, medium, high, very_high), with 1516 scenarios in the low tier, 869 in medium, 886 in high, and 1290 in very_high. Experiments in the paper use the first ten scenarios of the low tier. Data is provided in JSONL format, with each scenario object including fields such as product_name, product_description, buyer_persona, seller_persona, seller_res_price_range, and buyer_res_price_range.




