This thesis provides insight into long-term factors of user behaviour with a Web site or application using low-level interaction events (such as mouse movement, and scroll action) as a proxy. Current laboratory studies employ scenarios where confounding variables can be controlled. Unfortunately, these scenarios are not naturalistic or ecologically valid. Existing remote alternatives fail to provide either the required granularity or the necessary naturalistic aspect. Without appropriate longitudinal approaches, the effects of long-term factors can only be analysed via cross-sectional studies, ignoring within-subject variability. Using a naturalistic remote interaction data capturing tool represents a key improvement and supports the analysis of longitudinal user interaction in the wild. Naturalistic low-level fine-grained Web interaction data (from URLs visited, to keystrokes and mouse movements) has been captured in the wild from publicly available working live sites for over 16 months. Different combinations of low-level indicators are characterised as micro behaviours to enable the analysis of interaction captured for extended periods of time. The extraction of micro behaviours provides an extensible technique to obtain meaning from long-term low-level interaction data. 18 thousand recurring users have been extracted and 53 million events have been analysed. A relation of users' interaction time with the site and their degree of familiarity has been found via a remote survey. This relation enables the use of users' active time with the site as a proxy for their degree of familiarity. Analysing the evolution of extracted micro behaviours enables an understanding of how users' interaction behaviour changes over time. The results demonstrate that monitoring micro behaviours offers a simple and easily extensible post hoc approach to understand how Web-based behaviour changes over time. Results of the analysis have identified key aspects from micro behaviours that are strongly correlated with users' degree of familiarity. In the case of users scrolling continuously for short periods of time, it has been found that the speed of the scroll increased as users' become more familiar with the Web site. Users have also been found to spend more time on the Web site without interacting with the mouse. Understanding long-term interaction factors such as familiarity supports the design of interfaces that accommodate users' interaction evolution. Combining found key aspects enables a prediction of a user's degree of familiarity without the need for continuous observation. The presented approach also allows for the validation of hypothesis on longitudinal user interaction behaviour factors.