Modifier Graph Based Subtopic Mining

Given a role-explicit topic, we firstly determine its kernel-object, then we construct the corresponding modifier graph (Step-1). We perform graph clustering on the induced modifier graph using the algorithm by Noack [4]. Each modifier cluster is viewed as a representation of a particular subtopic. Corresponding to each modifier cluster, we generate the clusters of subtopic strings based on the “child-parent” relationship between modifiers and subtopic strings. For example, the middle part of the Figure 2 shows the clustered modifier graph of an official topic. The modifier clusters are differentiated by colors. The clusters of subtopic strings have the same color with their child modifiers. Based on the clusters of subtopic strings, we generate the target ranked list. Results of TUTA1