The CLIIP temporal learning framework has two inputs. The first one is continuous spatial data for building an IDG, and the other is a set of labels that provide people’s infected states. Combining interaction data with the states and the relation, we can train this model to learn continuously when new data comes into the framework. The updating IDG is built through comparing the place where two people stop and their overlap time, which defines the relation between two people. An arrow points to the person who stayed longer at a waypoint than the other. According to the path of virus transmission29, people’s continuous spatial data are a set of essential interaction features, \(X_i, i=1,2,3...\), which we use from mobile location data and we can call interaction data.