Catchphrase Generation Based on Strings Representing Personas of Target Consumers

Abstract

Generative AI, particularly Large Language Models (LLM), holds promise for natural language processing tasks such as catchphrase generation. However, the suitable methods aligning catchphrase generation by LLM with statistical data analysis are currently under investigation. This study proposes a method for automated catchphrase generation by LLM with persona strings derived from data analysis. Consumers are classified into groups based on their answers to questionnaires, and advertising effects are estimated using marketing survey data. After identifying consumer groups showing high advertising effects, personas strings representing them are extracted to generate catchphrases. The proposed method was evaluated by cosine similarity calculated with BERT between catchphrases generated by LLM with and without incorporating consumer persona strings. Results acquired from a dataset of 2500 marketing survey samples demonstrated that catchphrases generated with the inclusion of persona strings exhibited a higher similarity (by 0.02) compared to those generated without the strings.