The efficacy of a TCM medication derives from the herb-herb interaction in a formula. Although there are standard formulae, a practitioner will only pick a subset of formulas as templates and personalize them for the patients. It is not easy to determine the true interacting herbs to contribute to the effectiveness of a treatment. Association rule mining is an approach to find the co-occurrence of some items, however, it is not goal-oriented, and the generated results are very sensitive to the given parameters, i.e. support count. The aim of this paper is to introduce a new framework to systematically generate a set of combinations of interacting herbs that leads to good outcome. This algorithm was tested with a dataset of treatment of insomnia to understand the effectiveness of combination of herbs. Interesting and insightful results were noted and discussed.