NAMeGEn: Creative Name Generation via A Novel Agent-based Multiple Personalized Goal Enhancement Framework

TL;DR

The paper tackles creative natural language generation under hybrid multi-objective optimization for short-form tasks, exemplified by Chinese baby naming. It introduces NAMeGEn, a training-free multi-agent framework (MOM, MOG, MOE) that solves $Y = \arg\max_{\theta} f(x, O_{exp}, O_{imp})$ through two-stage information preparation and dynamic iterative optimization, including retrieval-augmented generation. The authors build the CBNames benchmark and the CPoetry corpus and demonstrate that NAMeGEn outperforms six baselines across multiple LLM backbones, achieving superior balance among explicit user objectives and implicit interpretive objectives while improving interpretability and reducing hallucinations. Overall, NAMeGEn offers robust cross-model performance and practical applicability for real-world CNLG tasks, enabling flexible, explainable creativity in constrained naming design.

Abstract

Trained on diverse human-authored texts, Large Language Models (LLMs) unlocked the potential for Creative Natural Language Generation (CNLG), benefiting various applications like advertising and storytelling. Nevertheless, CNLG still remains difficult due to two main challenges. (1) Multi-objective flexibility: user requirements are often personalized, fine-grained, and pluralistic, which LLMs struggle to satisfy simultaneously; (2) Interpretive complexity: beyond generation, creativity also involves understanding and interpreting implicit meaning to enhance users' perception. These challenges significantly limit current methods, especially in short-form text generation, in generating creative and insightful content. To address this, we focus on Chinese baby naming, a representative short-form CNLG task requiring adherence to explicit user constraints (e.g., length, semantics, anthroponymy) while offering meaningful aesthetic explanations. We propose NAMeGEn, a novel multi-agent optimization framework that iteratively alternates between objective extraction, name generation, and evaluation to meet diverse requirements and generate accurate explanations. To support this task, we further construct a classical Chinese poetry corpus with 17k+ poems to enhance aesthetics, and introduce CBNames, a new benchmark with tailored metrics. Extensive experiments demonstrate that NAMeGEn effectively generates creative names that meet diverse, personalized requirements while providing meaningful explanations, outperforming six baseline methods spanning various LLM backbones without any training.

Figures (9)

• Figure 1: Example of Chinese Baby Naming (NCB). Different colors indicate diverse objectives. The model seeks an optimal balance among them.
• Figure 2: Overview of NAMeGEn. Steps 1.1 and 1.2 constitute the multi-objective information preparation process, which is primarily handled by MOM and MOE. The dynamic iterative objective optimization process includes Steps 2 and 3: Step 2 is managed by MOG, while Step 3 reflects MOE’s role in evaluating the generation results. The green block at the bottom right illustrates the complete pipeline of NAMeGEn.
• Figure 3: Comparison of balance performance on EUOs, IIOs, and their overall combination across different methods. Colored lines correspond to different backbone models. We use method abbreviations for clarity: B = Base, C = CoT, T = TDB, Fs = Few-shot, Q = Q2Kw, LD = LLM-D, NG = NAMeGEn.
• Figure 4: Fine-grained comparison of explicit and implicit objective completeness. (a) shows explicit scores (blue = high, red = low) across five aspects. O1–O5 represent cultural significance, parental expectations, Bazi & Wuxing, personal traits, and other requirements. (b) shows implicit results, using different colors to indicate various backbones, and shade depth reflects different metrics. Method abbreviations are the same as in Fig. \ref{['fig_comp_std']}.
• Figure 5: Pairwise IIO comparisons. Blue lines show Pareto front; red line marks 80. Colors and shapes represent backbones and methods.