SYNTH

该数据集名为SYNTH，包含16,384个实体，它们被明确标记为两个截然不同的类别，并通过使用Watts-Strogatz算法生成的随机图结构相互连接。每个实体配备了不同数据类型的字面量，覆盖了五种模态。此外，该数据集缺乏关系信息，但被设计为具有强烈的多模态信号。字面值从两个狭窄且略有重叠的分布中抽取，并在必要时添加噪声。数据集的规模为16,384个实体，其任务包括节点分类和链接预测。

This dataset, named SYNTH, contains 16,384 entities explicitly labeled into two distinct categories and interconnected via a random graph structure generated using the Watts-Strogatz algorithm. Each entity is equipped with literals of various data types spanning five modalities. Furthermore, this dataset lacks relational information but is designed to exhibit strong multimodal signals. The literals are sampled from two narrow and slightly overlapping distributions, with noise added when necessary. The dataset has 16,384 entities in total, and its supported tasks include node classification and link prediction.