This document proposes extensions to user behavior modeling for web application prefetching. It discusses using n-gram and n-gram+ techniques to predict the next actions users will take based on sequential patterns in their historical requests and responses. Relations between actions are defined to identify dependencies between tokens in requests. An algorithm is proposed to assign actions to endpoints, tokenize requests/responses, identify action relations through n-gram statistics, and predict/prefetch future actions by filling token values. This predictive modeling could help prefetch dependent resources to reduce latency.