软件导刊2025，Vol.24Issue(5)：53-61,9.DOI:10.11907/rjdk.241226
基于深层个性信息的个性化对话生成技术
Technology for Generating Personalized Dialogues Based on Deep Persona Information
摘要
Abstract
In daily communication,people often unconsciously reveal their personality information during conversations.The personalized dia-logue system simulates this process by introducing personalized information into the model to generate personalized responses.Personalized di-alogue involves three core actions:personality expression,personality perception,and personality understanding.Personality expression re-fers to expressing one's own personality information in dialogue;Personality perception refers to inferring the personality information contained in the other person's words;Personality understanding refers to the classification and summary of personality information.Dialogue models equipped with these abilities can engage in more attractive conversations with humans.However,existing research only focuses on the ability to express personality in models,while neglecting the study of personality perception and understanding.To this end,a personalized dialogue dataset PPU(Persona Perception and Understanding)is proposed,which includes annotated information on individual perception and under-standing.Simultaneously design a personalized dialogue model PUT(Perception and Understanding Transformer)that can utilize deep person-ality information,including modules related to personality perception,understanding,and expression.In the experiments on the PersonaChat dataset,PUT reduced the perplexity index by an average of 5.39 compared to the baseline model,and improved the F1 index by an average of 1.86.Meanwhile,in the manually evaluated indicators of fluency and personality consistency,PUT showed an average improvement of 15.55%and 6.27%compared to the baseline model,respectively.The experimental results indicate that personality perception and under-standing can effectively enhance the personalized dialogue performance of the model.关键词
个性化对话/个性表达/个性感知/个性理解Key words
personalized dialogue/personality expression/personality perception/personality understanding分类
信息技术与安全科学引用本文复制引用
陈杰,蒋玉茹,区可明,佘泽鹏,张志华,秦晓博,王若凡,张仰森..基于深层个性信息的个性化对话生成技术[J].软件导刊,2025,24(5):53-61,9.基金项目
北京市自然科学基金项目(4242019) （4242019）
