Predicting CTR of Regional Flyer Images Using CNN

Abstract

In recent years, a variety of social media have been launched for regional revitalization. A Japanese social media service "Tame-map" is the one of them. Tame-map is a Web application for sharing local information, allowing users to easily post and view information about local events in their daily lives. Since many images are posted on Tame-map every day, creating a well-designed flyer which can attract many users is important. The more users view the site, the more participation in events can be expected. Therefore, increasing the number of views is an important issue from the perspective of revitalizing the local community. The more users view the site, the more participation in events can be expected. On the other hand, the methodology for creating a design that attracts many visitors has not been established. The design process completely depends on organizer’s experience and intuition. In this regard, if the number of views can be predicted to some extent in advance when posting information, the predicted views would be helpful to reconsider the design of the flyer. As a result, we can expect an increase in the number of advertising images that can be viewed by many users, which will lead to the revitalization of the local community. Then, in this study, we predict CTR (Click Through Rate) of local even flyers for the data of Tame-map. We construct a CTR prediction model using CNN, and show that the constructed model is useful for predicting the CTR