This research focuses on the problem of anomaly level calculation in the behavior of the user of an electronic information system. This type of problem is important for a variety of electronic systems that provide access to financial, medical, military and similar sensitive data. Such systems are well protected against external threats; however monitoring of internal, already authorized users, is less effective at the moment. To solve the problem, an approach is implemented that allows specifics of interests of the users to be represented as a graph of the Markov chain. Having this self−learning Behavioral Profile of the User (UBP), it is possible to calculate the level of abnormality of each its request to the system. A method for improving the accuracy of calculating the level of anomalous behavior through using a Personal Adaptive UBP (PAUBP) is developed. A method that uses a dynamic threshold for the anomaly level in order to increase the efficiency of classification between the "normal" and "abnormal" behavior is elaborated. A method of estimating the possibility of introducing the approach developed into the information systems having different structures is established. An approach to evaluate the effectiveness of UBP is proposed. A software system that enables making experiments with all the required characteristics of UBP is created. The experimental results show that PAUBP is more effective in solving the considered problem than the general UBP.