Abstract
The trajectory of a user's continuous online access, which manifests as a sequence of dynamic behaviours during online purchases, constitutes fundamental behavioural data. However, a comprehensive computational method for measuring trajectory similarity and thoroughly analyzing user behaviour remains elusive. Analyzing user behaviour sequences requires balancing detail with data reduction while addressing challenges such as excessive spatial complexity and potential null results in predictions. This study addresses two critical aspects: First, it evaluates similarity in the time dimension of user behaviour sequence clustering. Second, it introduces a frequent sub-trajectory mining algorithm that emphasizes the order of user visits for trajectory analysis and prediction. We employ a variable-order Markov model to manage the growth of probability matrix size in forecasts. Additionally, we improve prediction accuracy by aggregating the time spent on specific web pages.