Open-domain dialogue systems aim to chat with users in a human-like style, reaching considerable focus from academic and industrial circles. The existing studies mainly focus on the personality of the dialogue agent while ignoring the consistent expression. The dialogue systems with consistent expressing style can effectively acquire users’ trust and give full play to its value. To generate a consistent response, in this paper, we propose a personalized dialogue model named PAGE, which excavates and understand the characters immensely. Then, we use the fine-grained fusion of dialogue context and persona to improve generation quality. Finally, we utilize the language inference data and employ the unlikelihood training process to minimize the contradictory response with the persona. The experimental results demonstrate that our proposed model outperforms the baseline and the effectiveness for improving response consistency.