Network big data refers to the big data generated by the interaction and fusion of "people, machines and things" in the network space and available on the Internet. In this paper, from the perspective of ontology-based knowledge representation, network knowledge is represented in the form of five-tuples such as instance, attribute, domain set, relation and concept. Then, from the perspective of complex system, the new research perspective of deep learning technology and the new technology trends of grain computing in information fusion are analyzed, and the feasibility of the integration and expansion of grain computing and deep learning is speculated. A new overlapping community detection algorithm based on modularity optimization is designed. It is found that the framework can solve the dynamic update of network big data and the integration and reorganization of existing professional knowledge base to a certain extent. It can effectively guarantee the construction of domain knowledge base and improve its coverage and timeliness.