[4Rin1] Interactive 2

[4Rin1-70] Keyword-based Text Generation for Internet Advertisement

〇Kohei Wakimoto¹, Shunyo Kawamoto¹, Peinan Zhang¹ (1.CyberAgent, Inc.)

Keywords:Text Generation, Advertisement Generation, Natural Language Processing

Generating advertisements massively has a great demand for internet advertising. Banner ads, which consist of images and texts, are playing an important role that attracts users' interest, let them know about the product's information and encourage them to visit product's landing page (LP). Handcrafting a large number of ads, especially the texts, based on information on the LP is extremely expensive. Therefore, we attempted to generate these texts automatically. The generated ad texts must follow the description of the product described in the LP. However, using the raw LP for machine learning is challenging because both of the quality and the quantity of descriptions are different. Moreover, LP contains many noisy tokens that will interfere with the learning. In this paper, we propose a keyword-based generation of ad texts that will consider the information about the product. We experimented using the actual advertisement data and confirmed that this method will generate appropriate ad texts.