Personalized Emotional Intelligence in Generative AI through Symbolic Affective Reasoning

Jul 16, 2026

Upload by Rowan

Emotional intelligence helps people recognize emotions, infer their causes, reason about interventions, and reshape environments to reach desired affective states. Yet current AI models mainly generate realistic content or do semantic reasoning, with limited ability to understand, predict, and personalize human emotional responses. This work presents EROS, a hybrid framework combining symbolic reasoning and deep learning for personalized emotion augmentation via visual editing. Using large image–emotion datasets, EROS learns generalizable affective rules and predicts context-aware visual modifications while preserving scene semantics, supported by an inference-time memory bank for individual personalization and interpretable emotional profiles. Extensive psychophysics experiments show stronger target-emotion elicitation and faster adaptation than state-of-the-art multimodal models, with applications in mental health, adaptive media, education, and HCI.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages59

Reading Time2 h 29 min

Abstract

1 Introduction

Personalized affective image editing

Emotion-augmented geneRatiOn System (EROS)

Key challenges in affective understanding

arXiv :2607 . 10678v 1 [ cs .AI] 12 Jul 2026 Personalized Emotional Intelligence in Generative AI through Symbolic Affective Reasoning Qing Lin, Mengmi Zhang College of Computing and Data Science, Nanyang Technological University, Singapore, Address correspondence [to: mengmi.zhang@ntu.edu.sg](to: mengmi.zhang@ntu.edu.sg) Text statistics 6 figures 9 supplementary figures Abstract Emotional intelligence enables humans to recognize emotions, infer their causes, reason about interventions, and modify their environment to achieve desired affective states. Despite recent advances in artificial intelligence (AI), current models remain largely limited to generating realistic content or performing semantic reasoning, with little capacity for understanding, predicting, and personalizing human emotional responses. Here we introduce Emotion-augmented geneRatiOn System (EROS), a hybrid AI framework that integrates symbolic reasoning with deep learning to enable personalized emotion augmentation through visual content. Leveraging large-scale image–emotion datasets, EROS discovers generalizable affective rules, identifies emotion-relevant image regions, and predicts context-aware visual modifications that preserve scene semantics while steering emotional responses toward desired targets. To account for individual variability, EROS incorporates an expandable memory bank that supports inference-time personalization without model fine-tuning, yielding interpretable emotional profiles and rapid adaptation to new users. Across extensive human psychophysics experiments, EROS elicits target emotional responses more effectively than state-of-the-art large multimodal models while adapting to individual affective preferences. Beyond affective computing, EROS provides a foundation for AI systems that can understand, reason about, and augment human cognitive states, with potential applications in mental health, adaptive media, education, and human–computer interaction. 1 Introduction When communicating with a homesick friend, one may selectively emphasize warm lighting, familiar objects, or comforting visual cues in a photograph to evoke a sense of belonging (Fig.1A) . This ability to infer another person’s emotional state, understand how visual elements shape affect, reason about potential interventions, and modify visual content to achieve a desired emotional outcome is a hallmark of human emotional intelligence [1]–[4] (Fig.1B) . Computationally instantiating such emotional intelligence in machines remains a fundamental challenge. Emotional responses do not arise from individual objects in isolation but emerge from the holistic interpretation of scenes and their contextual relationships. In the example above (Fig.1A), lighting, familiar objects, spatial layout, and implied context jointly contribute to the perceived emotion, requiring models to reason about interactions among visual elements and their meanings [5]–[11] . Moreover, emotional perception is inherently personalized. The same image may evoke comfort and warmth for one observer but loneliness, fear, or indifference for another, reflecting differences in personal experience, cultural background, and affective traits [12]–[14] . Finally, emotional regulation through image modification is fundamentally underconstrained: many possible edits can induce similar emotional outcomes, yet only a subset preserves the semantic content and structural integrity of the original scene [15]–[17] . Recent advances in AI have enabled models to generate photorealistic images [18]–[20], learn increasingly powerful visual representations for recognition and localization [21]–[35], and support scene understanding and structured visual reasoning [36]–[41] . Despite these achievements, contemporary AI systems remain largely optimized for semantic understanding rather than affective understanding. They can recognize what is present in an image, but have limited ability to explain why visu

FAQs

What limitation in current AI motivates the proposed approach?

Existing models largely support realistic generation and semantic reasoning, but they struggle to understand why images evoke specific emotions, predict individual differences, and generate reliable context-aware edits for targeted emotional outcomes.

How does EROS achieve personalized emotional augmentation?

EROS combines symbolic reasoning with deep learning to learn compositional affective rules from large image–emotion datasets, then uses an explicit memory bank for inference-time personalization without model fine-tuning.

What is the goal of personalized affective image editing described in the paper?

Given a source image and a target emotional valence, the method generates minimally modified images that reliably evoke the desired emotion in a specific observer while preserving the original scene’s semantic meaning and structural consistency.

Jul 16, 2026

Upload by Rowan