Controllable Emotion Generation with Emotion Vectors

2025-02-06

AI Summary

Current large language models (LLMs) lack precise controllability over affective expression—specifically, tone, timing, and directness/indirectness. To address this, we propose a controllable affective generation method grounded in learnable emotion vectors. Our approach introduces, for the first time, a general-purpose, flexible, plug-and-play emotion vector interface enabling zero-shot affective control across diverse model scales and architectures—without fine-tuning. The method integrates emotion vector embedding, structured prompt engineering, a multi-dimensional emotion annotation and evaluation framework, and a few-shot controllable decoding strategy. We validate its effectiveness across multiple mainstream LLMs—including variants spanning different parameter counts and architectural families—demonstrating significant improvements in emotional accuracy, naturalness, and fine-grained controllability. The framework has been successfully deployed in real-world applications, including intelligent customer service systems and companion robots.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In recent years, technologies based on large-scale language models (LLMs) have made remarkable progress in many fields, especially in customer service, content creation, and embodied intelligence, showing broad application potential. However, The LLM's ability to express emotions with proper tone, timing, and in both direct and indirect forms is still insufficient but significant. Few works have studied on how to build the controlable emotional expression capability of LLMs. In this work, we propose a method for emotion expression output by LLMs, which is universal, highly flexible, and well controllable proved with the extensive experiments and verifications. This method has broad application prospects in fields involving emotions output by LLMs, such as intelligent customer service, literary creation, and home companion robots. The extensive experiments on various LLMs with different model-scales and architectures prove the versatility and the effectiveness of the proposed method.

Problem

• Enhance emotion expression in LLMs
• Control emotional tone and timing
• Apply to customer service and robotics

Innovation

• Emotion Vectors for LLMs
• Universal Emotional Expression Control
• Versatile Application in AI Services