API recommendation is an effective way to improve software productivity. It has become an important research topic in recent years, and therefore attracted lots of attention. Recently, researchers have proposed a great deal of API recommendation approaches to help developers find suitable APIs from a large number of candidates during the development process. However, most of these approaches do not effectively integrate dynamic human-machine interaction information into the recommendation session. In this article, we propose an API recommendation approach, ARENA (API Recommendation using intEractive reiNforcement leArning), which leverages reinforcement learning based on markov decision process (MDP) to improve the recommendation performance. By integrating the user interaction into the loop of MDP, ARENA enables users to continuously interact with the model. In this way, the user interaction information is used for optimizing the parameter of the model. In turn, the approach can provide personalized API recommendation result to end users.