Existing Chinese educational corpora often suffer from fragmented coverage and lack of systematic organization. To address these issues, we introduce WoW (World Knowledge), a large‑scale dataset designed for the full‑cycle training of foundation models. WoW is built on an Educational Knowledge Graph that serves as a taxonomic backbone, enabling systematic collection and completion of missing educational modules across all levels. Compared with previous versions, this iteration offers improved systematicity, higher quality, larger scale, and incorporates multimodal signals to support every stage of model development. Experiments show that models trained on WoW achieve significant gains on diverse benchmarks. To foster community research, we also provide an interactive platform for graph‑based data selection and retrieval.
Review: WoW’s graph‑driven, systematic construction delivers a more complete and higher‑quality training resource for Chinese educational LLMs. The inclusion of multimodal data and support for the entire development lifecycle make it a valuable foundation for future model innovations.