OpenNE is an open-source network embedding toolkit from Tsinghua University’s THUNLP lab, built for network representation learning. With about 1,700 stars on GitHub, it provides a standard training and testing toolkit that unifies many classic graph embedding models behind common input and output interfaces. The project was last updated in 2024 and remains a reference implementation for the field.
Network embedding learns low-dimensional vector representations of nodes in a graph, which can then feed downstream tasks such as node classification, link prediction and visualization. OpenNE implements a wide range of well-known models โ including DeepWalk, LINE, GraRep, SDNE and TADW, which adds text attributes โ and evaluates them under shared experimental settings so results are comparable across papers. It ships with benchmark datasets such as Cora and Wiki, plus scalable options for each component, making it a practical starting point for experiments in graph machine learning. Its maintainers come from the same lab behind several widely used NLP and knowledge-graph toolkits.
OpenNE is completely free and open source under the MIT license. There is no commercial edition or hosted service; you download the code and run experiments on your own hardware. As with most research toolkits, support comes from the community and the repository’s issue tracker.
Researchers and students working on graph representation learning, and ML engineers who need proven network embedding baselines for tasks like node classification and link prediction. It is best suited to users comfortable with Python research code rather than a polished end-user application.