Nowadays, knowledge based innovation design has been an essential technique to enhance enterprise's competitiveness. Nowadays, the main method to realize the innovation design is Structured Innovation (such as TRIZ) and Heuristic Innovation (such as biological inspired design). The former prefer to address specific issues through processes and tools, the latter is used to stimulate people's creativity, but how to provide the right information to designers? The problem is complex and it has become a research hotspot. Therefore, this paper proposed a methodology which is used to find the useful information from cross domain knowledge and then recommend the information to designers. In detail, we collect and process different domain literatures first, and then represent the data in matrix models. Finally, K-means clustering analysis is used to detect outlier documents, which will be recommended to designers. At last, a JAVA system-product innovation design supporting system was developed to validate the methodology.