DeepKE is a deep learning toolkit for knowledge graph construction that extracts entities, relations, and attributes from text.
The toolkit addresses the challenge of building knowledge graphs by providing specialized extraction capabilities across multiple scenarios. It supports entity recognition, relation extraction, and attribute extraction through supervised deep learning models. The approach accommodates low-resource settings, document-level extraction, multimodal inputs, and Chinese language processing via cnSchema. The toolkit includes implementations of several published models for named entity recognition, relation extraction, and relational triple extraction, allowing users to train supervised models or leverage pre-trained variants.
Teams building knowledge graphs should consider DeepKE if they need flexible extraction across entity, relation, and attribute tasks with support for Chinese text and low-resource scenarios. The toolkit suits projects requiring document-level understanding rather than sentence-level extraction alone. For those preferring large language model approaches, the toolkit provides integration options through DeepKE-LLM and OneKE. The project offers pre-trained models through DeepKE-cnSchema for immediate deployment, reducing the need for custom training in some cases.
Development activity shows consistent expansion of capabilities with multiple extraction model variants integrated into the toolkit. The project maintains comprehensive documentation alongside the codebase, including an online demonstration environment. Integration with large language models has been added as a parallel capability rather than replacing the supervised model approach. The toolkit provides troubleshooting guidance and actively addresses installation issues reported by users.
