The latest release of zjunlp/deepke published on GitHub is 2.2.7. It is a deep learning-based knowledge extraction toolkit written in Python and distributed under the MIT license. However, the record provided here contains no release notes or change list for 2.2.7, so what is new in this version cannot be confirmed.
The toolkit targets extracting the building blocks of knowledge graphs from unstructured text. A knowledge graph is a structure that connects knowledge with nodes and edges. The project description calls it an open toolkit for knowledge graph extraction and construction. It supports entity, relation, attribute, and event extraction, and the README describes entity, relation, and attribute extraction in cnSchema, low-resource, document-level, and multimodal scenarios. Each task keeps its own data processing, model, and training/evaluation/prediction components, and the same README also introduces DeepKE-LLM and OneKE, which make use of large language models (LLMs). The repository was created in August 2018, the description field mentions EMNLP 2022, and the GitHub star count is recorded as 4486.
The evidence stops at the release notes. The release date of 2.2.7, its change history, the download count, recent issue activity, and the reference point for the star count are not in this record. No quantitative figures or benchmarks indicating extraction accuracy are presented either. What can be confirmed is only the feature set described in the README and the MIT license. So to judge what 2.2.7 improved, the release notes have to be checked directly.