Abstract: Software business process needs higher flexibility and adaptability. Traditionally, various methods are developed to improve the flexibility of process during the modeling period, and the running period is omitted. It is hard to describe the environment model for its dynamic, thus, environment can't be controlled during the modeling period. Because the change of environment can be explained by the data of running period, analyzing the treated data of running period of process is necessary. We built a learning algorithm based on reinforcement learning, a well-known learning technique, to learn flexible constraints from the treated data of running period, and the adaptability of process is improved. We will evaluate our learning algorithm in a user-centred complex information system, and show that our algorithm is effective and efficient.

Key words: Reinforcement learning, Flexible constraint, Business process