aida-ugent/llm-ideology-analysis

该数据集包含来自17种不同大型语言模型对4,339位政治人物的评估，评估结果以英语和中文两种语言呈现。评估过程采用了两阶段的提示策略，旨在分析AI语言模型中的意识形态偏见。数据集的创建目的是量化LLM的意识形态立场，比较不同语言和文化背景下的偏见，并提供AI偏见和治理讨论的实证依据。数据集的结构包括第一阶段和第二阶段的响应、政治人物的元数据以及模型信息。数据集的来源包括Pantheon数据集、Wikipedia摘要和Manifesto项目分类法。数据集的使用范围明确，主要用于分析LLM中的意识形态偏见，而不应用于非意识形态任务的评估。

This dataset contains evaluations of 4,339 political figures by 17 different Large Language Models, with responses collected in both English and Chinese. The evaluations were conducted using a two-stage prompting strategy to analyze ideological biases in AI language models. The dataset was created to quantify the ideological positions of LLMs, compare biases across different languages and cultural contexts, and provide empirical evidence for discussions about AI bias and governance. The dataset structure includes stage 1 and stage 2 responses, political figure metadata, and model information. The dataset sources include the Pantheon dataset, Wikipedia summaries, and the Manifesto Project taxonomy. The dataset is intended for analyzing ideological biases in LLMs and is not to be used for evaluating non-ideological tasks.