Mining user behavior patterns based on reinforcement learning algorithm to optimize service robot interaction strategy

In order to resolve the challenges of reduced adaptability and generalization stemming from complicated user behavior and sparse feedback in service robot interactions, this paper presents a reinforcement learning-based approach to user behavior pattern mining and policy optimization. This approach integrates Bayesian belief updates and automata learning with counterexamples to unify intent modeling and policy iteration: dynamic intent reasoning promotes multi-scale exploration under sparse rewards; and counterexample reasoning reconstructs the rewards function to bolster policy generalization. Experiments revealed that the method led to mean latency standard deviation of 0.031, a task completion rate was 78.14%, and behavior recognition accuracy 0.9, which can contribute to more capable policies and improve user satisfaction, while providing a reference for the design of complex interaction strategies.