Approaches to detect fault-prone modules have been studied for a long time. As one of these approaches, the authors proposed a technique using a text filtering technique. They assume that bugs relate to words and context that are contained in a software module. The technique treats a module as text information. Based on the dictionary which was learned by classifying modules which induce bugs, the bug inducing probability over a target module is calculated, and it judges whether the given module is a fault-prone module. The predictive granularity of this technique is a module. In this study, the authors aimed at prediction with the finer granularity of the portion that induces a bug. Specifically, they tried to predict bug-inducing changes by using source code differences of bug inducing changes and previous changes and a text filtering technique. Similarly, the authors tried to predict bug fixing by using source code differences of bug fixing changes and previous changes and a text filtering technique. To show the effectiveness of the approach, the authors conducted two experiments and compared their approach with fault-prone filtering by applying it to two open source projects, and obtained higher accuracy.