Controllable Stylistic Text Generation with Train-Time Attribute-Regularized Diffusion

2025-10-07

AI Summary

To address low attribute control accuracy, high computational overhead, and poor generalization in controllable text generation, this paper proposes RegDiff—a novel framework that explicitly incorporates attribute regularization into the training phase of diffusion models. Unlike classifier-guided sampling, RegDiff jointly optimizes text reconstruction and attribute supervision in the latent space, eliminating reliance on external classifiers during inference. Built upon a VAE-based encoder-decoder architecture, it ensures semantic fidelity while enabling fine-grained attribute alignment within the latent diffusion process. Extensive experiments across five multi-attribute benchmark datasets demonstrate that RegDiff significantly outperforms strong baselines: it improves attribute control accuracy by 12.3%, enhances language quality (BLEU +4.1, BERTScore +2.8), and reduces sampling latency by 37%. These results validate RegDiff’s effectiveness, efficiency, and generalization capability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Generating stylistic text with specific attributes is a key problem in controllable text generation. Recently, diffusion models have emerged as a powerful paradigm for both visual and textual generation. Existing approaches can be broadly categorized into classifier-free guidance (CFG) and classifier guidance (CG) methods. While CFG effectively preserves semantic content, it often fails to provide effective attribute control. In contrast, CG modifies the denoising trajectory using classifier gradients, enabling better attribute alignment but incurring high computational costs during sampling and suffering from classifier generalization issues. In this work, we propose RegDiff, a regularized diffusion framework that leverages attribute features without requiring a pretrained classifier during sampling, thereby achieving controllable generation with reduced computational costs. Specifically, RegDiff employs a VAE-based encoder--decoder architecture to ensure reconstruction fidelity and a latent diffusion model trained with attribute supervision to enable controllable text generation. Attribute information is injected only during training. Experiments on five datasets spanning multiple stylistic attributes demonstrate that RegDiff outperforms strong baselines in generating stylistic texts. These results validate the effectiveness of RegDiff as an efficient solution for attribute-controllable text diffusion. Our code, datasets, and resources will be released upon publication at https://github.com/xxxx.

Problem

• Achieving controllable text generation with specific stylistic attributes
• Reducing computational costs in attribute-guided diffusion models
• Improving attribute alignment without classifier dependency during sampling

Innovation

• Train-time attribute regularization without classifier
• VAE encoder-decoder with latent diffusion model
• Attribute injection only during training phase