Relationship Between Diversity and Quality in Advertising-Text Generation

AACL-IJCNLP 2025 Workshop SRW ARR Commitment Submission3 Authors

27 Oct 2025 (modified: 31 Oct 2025)Submitted to IJCNLP-AACL 2025 SRW (ARR Commitment)EveryoneRevisionsBibTeXCC BY 4.0
Keywords: Generation, Analysis, Automatic Evaluation
Submission Category: Short Paper
Abstract: In natural language generation for advertising, diverse and engaging advertising (ad) texts must be created to capture a broad audience and avoid advertising fatigue. Despite the importance of diversity, the effects of diversity-enhancing methods in ad-text generation, which have mainly been tested on tasks such as summarization and machine translation, have not been explored thoroughly. Ad-text generation differs significantly from other tasks owing to the text style and requirements. Therefore, this study investigated the relationship between diversity and ad quality in Japanese ad-text generation by considering factors such as the diversity-enhancing method, hyperparameters, input--output format, and model.
Student Status Proof: pdf
Paper Link: https://openreview.net/forum?id=2jqnN10e51
Submission Number: 3