laurajul/SD_social

该数据集通过CivitAI API抓取了两年的图像数据，并通过分页机制和进度保存来优化抓取效率。数据抓取后，进行了标准化处理，包括时间惩罚和反应值归一化，以确保数据的时效性和可比性。时间惩罚通过应用基于内容在平台上存在时间的衰减来减少旧帖子的影响，而反应值归一化则将原始社交反应值缩放到标准化范围（0-1），以便进行比较。

The dataset was constructed by scraping image data over a two-year period using the CivitAI API, with optimizations for efficiency through pagination and progress saving. After data collection, normalization processes were applied, including time penalty and reaction normalization, to ensure the timeliness and comparability of the data. The time penalty reduces the influence of older posts by applying a decay based on how long the content has been on the platform, while reaction normalization scales raw social reaction values into a standardized range (0–1) for comparison.