Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generating Attractive and Authentic Copywriting from Customer Reviews

Yu-Xiang Lin, Wei-Yun Ma

TL;DR

This work tackles automatic, attractive, and authentic copywriting by deriving content from customer reviews rather than seller-provided attributes. It presents a reinforcement learning framework that combines supervised fine-tuning with three dedicated reward models—Allure, Veracity, and Information—to balance appeal, factual fidelity, and informative richness. Using a GPT-3.5-assisted dataset construction and a Yelp restaurant review corpus, the approach outperforms baselines including zero-shot LLMs and ROUGE-optimized models, as evidenced by human and automatic evaluations. The study demonstrates the practicality of producing marketing copy that is both persuasive and faithful to source reviews, with a publicly available codebase and dataset to support future research and deployment.

Abstract

The goal of product copywriting is to capture the interest of potential buyers by emphasizing the features of products through text descriptions. As e-commerce platforms offer a wide range of services, it's becoming essential to dynamically adjust the styles of these auto-generated descriptions. Typical approaches to copywriting generation often rely solely on specified product attributes, which may result in dull and repetitive content. To tackle this issue, we propose to generate copywriting based on customer reviews, as they provide firsthand practical experiences with products, offering a richer source of information than just product attributes. We have developed a sequence-to-sequence framework, enhanced with reinforcement learning, to produce copywriting that is attractive, authentic, and rich in information. Our framework outperforms all existing baseline and zero-shot large language models, including LLaMA-2-chat-7B and GPT-3.5, in terms of both attractiveness and faithfulness. Furthermore, this work features the use of LLMs for aspect-based summaries collection and argument allure assessment. Experiments demonstrate the effectiveness of using LLMs for marketing domain corpus construction. The code and the dataset is publicly available at: https://github.com/YuXiangLin1234/Copywriting-Generation.

Generating Attractive and Authentic Copywriting from Customer Reviews

TL;DR

This work tackles automatic, attractive, and authentic copywriting by deriving content from customer reviews rather than seller-provided attributes. It presents a reinforcement learning framework that combines supervised fine-tuning with three dedicated reward models—Allure, Veracity, and Information—to balance appeal, factual fidelity, and informative richness. Using a GPT-3.5-assisted dataset construction and a Yelp restaurant review corpus, the approach outperforms baselines including zero-shot LLMs and ROUGE-optimized models, as evidenced by human and automatic evaluations. The study demonstrates the practicality of producing marketing copy that is both persuasive and faithful to source reviews, with a publicly available codebase and dataset to support future research and deployment.

Abstract

The goal of product copywriting is to capture the interest of potential buyers by emphasizing the features of products through text descriptions. As e-commerce platforms offer a wide range of services, it's becoming essential to dynamically adjust the styles of these auto-generated descriptions. Typical approaches to copywriting generation often rely solely on specified product attributes, which may result in dull and repetitive content. To tackle this issue, we propose to generate copywriting based on customer reviews, as they provide firsthand practical experiences with products, offering a richer source of information than just product attributes. We have developed a sequence-to-sequence framework, enhanced with reinforcement learning, to produce copywriting that is attractive, authentic, and rich in information. Our framework outperforms all existing baseline and zero-shot large language models, including LLaMA-2-chat-7B and GPT-3.5, in terms of both attractiveness and faithfulness. Furthermore, this work features the use of LLMs for aspect-based summaries collection and argument allure assessment. Experiments demonstrate the effectiveness of using LLMs for marketing domain corpus construction. The code and the dataset is publicly available at: https://github.com/YuXiangLin1234/Copywriting-Generation.
Paper Structure (34 sections, 6 equations, 2 figures, 10 tables)

This paper contains 34 sections, 6 equations, 2 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Copywriting Generation
  4. Attractive Arguments Generation
  5. Aspect-based Summarization
  6. Dataset
  7. Methodology
  8. Supervised Fine-tuned Copywriter
  9. Reward Modeling
  10. Allure Reward
  11. Veracity Reward
  12. Information Reward
  13. Reinforcement Learning
  14. Experiments
  15. Allure Reward Model
  16. ...and 19 more sections

Figures (2)

  • Figure 1: The illustration for dataset construction. For simplicity, the requirements of the response format are skipped. We initially replaced "description" with "summary", resulting in a collection of monotonous text.
  • Figure 2: The illustration of the proposed framework. Given an aspect $k$, a review $x$, and a list of queries $F$, we enhance the supervised finetuned model $\theta$ by the Proximal Policy Optimization (PPO) algorithm with three dedicated reward models.