[3Yin2] Interactive session 1

[3Yin2-07] Advertisement text creation support using product reviews and features

〇Daisuke Taniguchi¹, Kohei Wakimoto², Kazuya Kuroda², Shunyo Kawamoto², Keisuke Nishiguchi², Ayana Niwa¹, Naoaki Okazaki¹ (1.Tokyo Institute of Technology, 2.CyberAgent, Inc.)

Keywords:Advertisement, Language model

In Internet advertising, it is necessary to produce many advertising messages for various consumers. Therefore, to understand the interests of the target consumers, ad creators sometimes devise expressions by referring to the feedback received from e-commerce sites and review sites when using the products. By automatically predicting feedback based on product features, it is possible to obtain more diverse expressions for a single product and support the creation of advertising messages that include expressions close to the consumer's viewpoint. Therefore, we built a model to generate feedback using GPT-2. We confirmed that the proposed method could generate desirable feedback to some extent through automatic and manual evaluation.