Personalized Emotional Support Conversation (ESC) systems (i.e., supporters) assist users (i.e., seekers) in navigating negative emotional states through personalized, empathetic interactions; these systems are often equipped with a persona extractor. Currently, personalized ESC systems face two key challenges. First, while existing persona extractors attempt to infer persona from dialogue to understand seekers, they often struggle to distinguish the speakers’ roles and thus only consider the utterances from seeker’s side. Second, incorporating personal information without consideration of contextual relevance risks damaging the naturalness and coherence of responses. Therefore, we present a novel Adaptive Persona Context Modulation approach (APCM) for the ESC task. For more effective persona extraction, we reconstruct the Persona-Chat dataset to adapt to our task and propose a role-cognitive persona extractor, thus enhancing the comprehensive understanding for the seeker’s persona from utterances by both sides while preventing role ambiguity. For persona-context integration, our model introduces an Adaptive Attention Balancing Module that dynamically adjusts the influence of persona and context information during response generation, better reflecting real-world conversation patterns where the seeker’s persona is only considered in appropriate circumstances. Extensive experiments on benchmark datasets demonstrate the effectiveness of APCM, achieving state-of-the-art (SOTA) performance in emotional support dialogue generation. Our code is available at https://github.com/hurricanewhk/APCM .