CS Knowledge Graph Dataset Family (CS-KG)

This repository contains the Computer Science Knowledge Graph (CS-KG) dataset family, a large-scale collection of heterogeneous knowledge graphs constructed from OpenAlex and Semantic Scholar metadata. The dataset captures relational structures among four key entity types—Papers, Authors, Venues, and Concepts—with five principal relation types: AUTHORED (Author → Paper, asymmetric) CITES (Paper → Paper, asymmetric and temporal) PUBLISHED_IN (Paper → Venue, functional) BELONGS_TO (Paper → Concept, hierarchical) COLLABORATES_WITH (Author ↔ Author, symmetric) Together, these relations form a multi-relational scholarly network well-suited for evaluating knowledge graph embeddings, relational learning, and geometric deep learning models. Each subgraph—CS-1K, CS-10K, CS-100K, CS-1M, and CS-10M—represents progressively larger scales (from thousands to millions of entities) and preserves real-world relational diversity across hierarchical, symmetric, and asymmetric patterns. All datasets have been preprocessed to remove duplicates and maintain temporal consistency by publication year.本仓库收录了计算机科学知识图谱（Computer Science Knowledge Graph，缩写CS-KG）数据集家族，这是一套依托OpenAlex与Semantic Scholar元数据构建的大规模异质知识图谱集合。该数据集涵盖四类核心实体——论文（Papers）、作者（Authors）、发表场所（Venues）与概念（Concepts）——之间的关系结构，共包含五种主要关系类型： 1. 撰写（AUTHORED，作者→论文，非对称关系） 2. 引用（CITES，论文→论文，非对称且带时间属性） 3. 发表于（PUBLISHED_IN，论文→发表场所，函数型关系） 4. 隶属于（BELONGS_TO，论文→概念，层级关系） 5. 合作（COLLABORATES_WITH，作者↔作者，对称关系） 上述五类关系共同构建了一个多关系学术网络，可广泛用于知识图谱嵌入、关系学习以及几何深度学习模型的性能评估。 其各子图——CS-1K、CS-10K、CS-100K、CS-1M与CS-10M——分别对应规模逐步递增的数据集（实体数量从数千到数百万不等），且完整保留了真实世界中层级、对称与非对称等各类关系模式的多样性。所有数据集均经过预处理，已剔除重复实体，并依据论文发表年份维持了时间一致性。