We propose a novel data mining framework using relaxed biclique for heterogeneous data. The framework is composed of three algorithms. First, an enumeration algorithm transforms heterogeneous databases into relaxed bicliques. Second, a tracking algorithm is used to find the biclique's variations over time. Finally, a ranking algorithm classifies relaxed bicliques into groups according to their statistical properties and dynamic behaviors. The framework is highly flexible and can be easily extended to applications in different domains. The framework is implemented in MapReduce and is proven to be scalable for processing large-scale data in a reasonable amount of time. In addition, the experiments show that the algorithms are both scalable and efficient. The proposed framework can also be applied to web network analysis and deliver rapid-response solutions.