Abstract : Human-computer interactions can be studied under several angles: some analyses attempt to describe human behavior using the computer as a tool, others seek to optimize the ergonomics of the interface, and finally, others seek to improve the interactivity between the user and the immense quantity of information conveyed by the numerical device. After having studied the properties necessary to design a generic system to gather data from human-computer interaction, we present methods of data visualization for the purpose of identifying useful properties, as well as examples of automatic analyses by algorithmic methods. We propose a new generic formalization of data harvesting, designed for data derived from human-computer interaction. This formalization allows all modeling, and all applications of all the methods previously exposed. We also present tools for analysis which allow automatic recognition of behavior characteristics of users, as well as a tool for data visualization. Application of the data-visualization tool permits us to replace the global interaction of the user in context as well as comparing the activity of several users on one or more interfaces. Two algorithms are introduced to facilitate the reading of this representation. As this work must be applicable in a scholastic context, we have studied various teaching theories elaborated by cognitivists. This has enabled us to develop a platform which allows the implementation of methods for machine learning and describes ways for adapting the interaction automatically to provide personalized help. Algorithmic methods of artificial intelligence are also classified by a typology of data processing. Two analyses carried out on data collected during experiments enabled us to work out methods that can be used for teaching by automatic personalization of human-computer interaction.