Abstract: The article is divided into two major parts, namely "collaborative innovation network classification" and "analysis of network innovation capabilities". The first part of the main solution is "What types of collaborative innovation networks?" and "How is the overall feature of each network type?". The second part is how the analysis factor affects network innovation performance? In order to solve the above problems, this paper is based on collaborative innovation perspectives. According to the research objectives and social network analysis theory, the common patent author of new energy automobile technology is the research object, first identify collaborative innovation networks with the Louvain community division algorithm, and calculate each network Network structural features such as network density, network scale, network clustering coefficient, and network average path length. According to the network structure characteristics, the K-Means clustering algorithm is used to find a collaborative innovation network with similar structures, and further analyze the overall characteristics of the network. Take the number of inventors holding patents, the number of IPCs, the number of patent citations, the number of claims, and the number of patent rights as basic indicators, and use the entropy method to comprehensively evaluate the quality of patents, and measure the performance of network innovation together with the number of patents. For each type of network, with network cooperation intensity and network structural characteristics, network innovation capability is a decision attribute, using CART decision tree to extract the conditions for decision attributes on decision attributes, answer "What factors affect each network type How it affects the problem”. From the perspective of data driven, the new method deeply analyzes the nonlinear relationship between the structural characteristics of collaborative innovation, cooperation intensity and innovation performance with the help of data mining technology. The results show that：(1) There are 4 types of the networks, namely the binary collaborative innovation network, star collaborative innovation network, completely collaborative innovation network and complex co-innovation network, different network types and its innovation ability is obvious, confirmed The rationality and scientificity of classification discussions. Binary cooperation innovation network cooperation relationship is simple; the star collaborative innovation network has obvious core members, non-core members have not produced cooperative relations; compared to the first three types, complex co-innovation network has more complex cooperative relationships. (2) Binary collaborative innovation networks are more cooperation with mother subsidiaries, and the network structure is simple; the non-core members of the star collaborative innovation network rely on core members, reduces the update speed of heterogeneity knowledge in the network. The innovation subjects of complete collaborative innovation networks are close; complex collaboration innovation networks are large, resulting in a complex partnership between collaborative subjects, and the membership relationship is relatively low. (3) Without the common influence of other factors, the cooperation intensity of the complex collaborative innovation network and the star-shaped collaborative innovation network have a positive and negative impact on the network innovation performance respectively; in the dual collaborative innovation network, the high or low The intensity of cooperation has a positive impact on the performance of network innovation. For collaborative innovation networks with large cooperation intensity, the average path length has a negative impact on network innovation capabilities.