On the use of cluster analysis for assisting maintenance of large software systems

With the ever-increasing size and complexity of software systems, their maintenance becomes a more and more difficult issue. Therefore, classical managerial solutions cannot be applied for maintaining very large software systems. The maintenance task must be assisted by automated techniques. Most existing tools can assist maintenance tasks only by requiring a lot of human-given information at the development stage. In contrast, the author claims that a lot of useful information can be directly retrieved from the existing code (and in the best cases, the natural language documentation) and that maintenance tools should be based on such automatically extracted information. The author performs a structural analysis of large software systems via adequate clustering techniques. This analysis allows retrieving useful information from the system that directs the considered maintenance task. Two tools that embody this approach in the domains of change management and reusability are also described