AdParaphrase: Paraphrase Dataset for Analyzing Linguistic Features toward Generating Attractive Ad Texts

TL;DR

AdParaphrase introduces a publicly available paraphrase dataset of semantically equivalent ad texts that differ in wording and style to analyze linguistic features driving human attractiveness judgments. The authors construct the dataset from Ad Similarity and CAMERA, annotate paraphrase relations, and collect attractiveness judgments with quality controls, revealing that fluency, length, noun density, noun phrases, and bracket usage correlate with attractiveness. Using chi-square analyses, they identify key linguistic predictors and demonstrate that prompting language models with these findings can improve ad-text attractiveness, though results in online pavings are mixed and room for improvement remains. The work offers a practical writing-assistant direction for ad production and highlights limitations related to language scope and dataset scale, suggesting multilingual expansion and broader feature exploration as future directions.

Abstract

Effective linguistic choices that attract potential customers play crucial roles in advertising success. This study aims to explore the linguistic features of ad texts that influence human preferences. Although the creation of attractive ad texts is an active area of research, progress in understanding the specific linguistic features that affect attractiveness is hindered by several obstacles. First, human preferences are complex and influenced by multiple factors, including their content, such as brand names, and their linguistic styles, making analysis challenging. Second, publicly available ad text datasets that include human preferences are lacking, such as ad performance metrics and human feedback, which reflect people's interests. To address these problems, we present AdParaphrase, a paraphrase dataset that contains human preferences for pairs of ad texts that are semantically equivalent but differ in terms of wording and style. This dataset allows for preference analysis that focuses on the differences in linguistic features. Our analysis revealed that ad texts preferred by human judges have higher fluency, longer length, more nouns, and use of bracket symbols. Furthermore, we demonstrate that an ad text-generation model that considers these findings significantly improves the attractiveness of a given text. The dataset is publicly available at: https://github.com/CyberAgentAILab/AdParaphrase.

Figures (9)

• Figure 1: Overview of human preference judgments for a pair of paraphrase ad texts.
• Figure 2: Distribution of Jaccard similarity for paraphrases and non-paraphrases in AdParaphrase.
• Figure 3: Distribution of maximum number of votes between ad text pair in attractiveness evaluation.
• Figure 4: Instruction for paraphrase creation presented to human experts.
• Figure 5: Prompt for paraphrase creation presented to LLMs.