Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GCOF: Self-iterative Text Generation for Copywriting Using Large Language Model

Jianghui Zhou, Ya Gao, Jie Liu, Xuemin Zhao, Zhaohua Yang, Yue Wu, Lirong Shi

TL;DR

The paper addresses the challenge of generating domain-specific marketing copy that reliably maximizes click-through rate (CTR). It presents the Genetic Copy Optimization Framework (GCOF), which fuses in-prompt feature engineering with a differential evolution-inspired crossover within a genetic algorithm, guided by a GPT-4-based reward model to iteratively refine copy. Key contributions include a novel integration of explicit prompt-level feature selection, a DE-informed keyword crossover mechanism, and an LLM-based reward scoring system that achieves substantial CTR gains (approximately $50\%$) over human-generated text. The work demonstrates a practical, scalable approach to marketing-copy optimization using large language models, with potential for cross-campaign knowledge transfer and reduced manual effort.

Abstract

Large language models(LLM) such as ChatGPT have substantially simplified the generation of marketing copy, yet producing content satisfying domain specific requirements, such as effectively engaging customers, remains a significant challenge. In this work, we introduce the Genetic Copy Optimization Framework (GCOF) designed to enhance both efficiency and engagememnt of marketing copy creation. We conduct explicit feature engineering within the prompts of LLM. Additionally, we modify the crossover operator in Genetic Algorithm (GA), integrating it into the GCOF to enable automatic feature engineering. This integration facilitates a self-iterative refinement of the marketing copy. Compared to human curated copy, Online results indicate that copy produced by our framework achieves an average increase in click-through rate (CTR) of over $50\%$.

GCOF: Self-iterative Text Generation for Copywriting Using Large Language Model

TL;DR

The paper addresses the challenge of generating domain-specific marketing copy that reliably maximizes click-through rate (CTR). It presents the Genetic Copy Optimization Framework (GCOF), which fuses in-prompt feature engineering with a differential evolution-inspired crossover within a genetic algorithm, guided by a GPT-4-based reward model to iteratively refine copy. Key contributions include a novel integration of explicit prompt-level feature selection, a DE-informed keyword crossover mechanism, and an LLM-based reward scoring system that achieves substantial CTR gains (approximately ) over human-generated text. The work demonstrates a practical, scalable approach to marketing-copy optimization using large language models, with potential for cross-campaign knowledge transfer and reduced manual effort.

Abstract

Large language models(LLM) such as ChatGPT have substantially simplified the generation of marketing copy, yet producing content satisfying domain specific requirements, such as effectively engaging customers, remains a significant challenge. In this work, we introduce the Genetic Copy Optimization Framework (GCOF) designed to enhance both efficiency and engagememnt of marketing copy creation. We conduct explicit feature engineering within the prompts of LLM. Additionally, we modify the crossover operator in Genetic Algorithm (GA), integrating it into the GCOF to enable automatic feature engineering. This integration facilitates a self-iterative refinement of the marketing copy. Compared to human curated copy, Online results indicate that copy produced by our framework achieves an average increase in click-through rate (CTR) of over .
Paper Structure (11 sections, 3 equations, 4 figures, 1 table)

This paper contains 11 sections, 3 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. GA in automatic feature engineering
  5. Reward Model
  6. Results and Discussions
  7. User Acquisition Through SMS
  8. User Acquisition Through Banner
  9. Conclusion
  10. Copy Generation Prompt
  11. Reward Model

Figures (4)

  • Figure 1: A marketing campaign card within JD Finance app. Promotional copy encompasses three components: entitlement, target audience and call-to-action.
  • Figure 2: The modified GA pipeline with GCOF Framework: After population initialization, we conduct keyword genetic crossover based on that within DE. The resulting keywords is then fed into LLM to produce the final copy. A reward model evaluates this output to assign a fitness score.
  • Figure 3: Schematic representation of the two-part prompt process for marketing copy generation using GCOF. The first panel illustrates the keyword combination generation through the Genetic Algorithm (GA), while the second panel depicts the subsequent utilization of these GA-derived keywords within the Prompt Framework to generate the final marketing copy.
  • Figure 4: Depiction of the prompt submitted to the Large Language Model (LLM) for generating a reward model, detailing the scoring criteria applied to evaluate a piece of marketing copy.