Highlight is employing Markov chains to identify and analyze interesting user sessions by examining the order of page visits and time spent on each page. To achieve this, the company has developed a method to define a "page" by normalizing URLs, removing resource IDs, and discarding intermediate, non-meaningful page visits. Using this normalized data, they calculate the probability of each state transition in a user journey, with less probable transitions indicating more interesting sessions. To address the bias towards longer sessions, Highlight normalizes these probabilities so that the expected value for each state transition remains consistent. This approach allows them to identify sessions with high interest by scoring and sorting them based on probability, with the aim of assisting users in uncovering unexpected behaviors in their applications. The methodology is being integrated into Highlight’s Session Insights email digest, which provides a weekly summary of intriguing user sessions.