This thesis is motivated by a case study of SSME, in which we digitalize and integrate cultural assets of TCM and provide Web-based knowledge services to medical experts. The major quest is to turn cultural assets from prolonged Chinese history into knowledge services contributing to modern biomedicine. In our view, the essence of knowledge service is cross-domain collaboration in knowledge discovery on the Web of data. Whereas the Service-Oriented Architecture enables interactions between Web agents, and the Semantic Web provides a knowledge representation and integration framework, the feasibility and benefits of Web-based collaborative knowledge discovery need to be further investigated. We propose a methodology named Semantic Graph Mining (SGM), which uses the semantic graph model to integrate graph mining and ontology reasoning for better analyzing biomedical complex networks (an important KDD problem). Potential methods of SGM include Web resource ranking, semantic association discovery, frequent subgraph mining, and clustering. The effectiveness of these methods is investigated in use cases such as TCM semantic search, TCM formulae analysis, drug-interaction analysis.