Information systems nowadays, ERP for instance, have almost always integrated user logging mechanisms. The user activities are logged in event logs which contain fields as activities’ names, timestamps and involved resources. On the grounds of these kinds of systems, Process Mining [1] aims first at discovering process models from event logs. However, the discovered process models are said to be activity oriented as they represent the activities and their ordering [2]. We believe activity oriented process models are useful when the processes are linear, mature, and do not involve creativity. Strategy oriented process models were proposed [3] to represent creative and flexible processes comprising the concepts of intentions (the objectives to be achieved), and strategies (the different manners to achieve the intentions). Those models are instances of a metamodel named Map, that has proved to be effective at modeling strategic alignment [4], defining actors and roles, specifying the outcome of business process models and naming them [5], supporting guidance [6], describing intentional services [7], expressing pervasive information systems [8], studying users' behavior to identify and name use cases, tailoring methods [9] or defining more flexible methods [9].Our research focuses on Intention mining: the discovery of intentions and strategies from event logs and their formalization in a map process model. Mining intentional process models allows a better understanding of the ways of working and thinking of systems users. The discovered intentional process model can be used as guidance for new users and be integrated into recommender systems to provide recommendations at runtime by analyzing the current intention of the user.We proposed a first approach based on Hidden Markov Models [11], [12] that allows discovering strategies from event logs and intentions by applying HMM. The naming of the strategies and intentions are done manually from the activities that are related to the strategies. This approach was proposed using supervised and unsupervised learning. The models obtained using unsupervised learning have proved to be more accurate. This approach has been validated using logs from Eclipse Usage Data Collector [13].We proposed another approach, FlexPAISSeer [14] that comprises two algorithms: IntentMiner which discovers the intentional model from event logs in an unsupervised manner, and IntentRecommender which generates recommendations as intentions and confidence factors, based on the mined intentional process model and probabilistic calculus.