To realize the benefits of Aspect Oriented Programming (AOP), developers must refactor active and legacy code bases into an AOP language. When refactoring, developers first need to identify refactoring candidates, a process called aspect mining. Humans perform mining by using a variety of clues to determine which code to refactor. However, existing approaches to automating the aspect mining process focus on developing analyses of a single program characteristic. Each analysis often finds only a subset of possible refactoring candidates and is unlikely to find candidates which humans find by combining analyses. In this paper, we present Timna, a framework for enabling the automatic combination of aspect mining analyses. The key insight is the use of machine learning to learn when to refactor, from vetted examples. Experimental evaluation of the cost-effectiveness of Timna in comparison to Fan-in, a leading aspect mining analysis, indicates that such a framework for automatically combining analyses is very promising.