Adaptive text generation with personality types and continuous emotion intensity

The goal of affective text generation is to produce language that aligns with specific emotional states, thereby enabling more natural and psychologically coherent human–computer interaction. However, existing approaches primarily focus on discrete emotion categories and often overlook the influence of stable personality traits on linguistic style and emotional expression. In addition, current prompting-based methods with large-scale language models lack fine-grained controllability over both personality and emotion intensity. To address these limitations, this study proposes a personality- and emotion-aware adaptive text generation framework for Chinese. The proposed approach jointly models personality types based on the Myers–Briggs Type Indicator (MBTI) and continuous emotional intensity across multiple emotion dimensions. A pre-trained generative language model is fine-tuned using structured prompts that incorporate input text, personality type, and emotion intensity. During inference, an adaptive decoding strategy dynamically adjusts sampling parameters, including temperature, nucleus sampling probability, and repetition penalty, through a sigmoid-based mapping mechanism guided by emotional intensity and personality attributes. To support this framework, a structured Chinese dataset is constructed by combining social media text with personality annotations and multi-dimensional emotion intensity labels. Experimental results demonstrate that the proposed method achieves improved performance in both stylistic consistency and emotional alignment compared to baseline approaches. The results highlight the effectiveness of jointly modeling personality and continuous emotion intensity for controllable text generation, with potential applications in intelligent dialogue systems, personalized content generation, and human-centered artificial intelligence systems.