Dentsu Inc. Develops "SHAREST (β version)", a TV Ratings Prediction System Using Deep Learning

The text of the Dentsu Inc. news release distributed on June 16 is as follows.

---

June 16, 2017

Dentsu Inc. Develops "SHAREST (β version)", a TV Viewership Prediction System Using Deep Learning

 

Dentsu Inc. (Headquarters: Minato-ku, Tokyo; President: Toshihiro Yamamoto) has developed "SHAREST (β version)" (Share Rest Beta), a system capable of highly accurate television ratings prediction using deep learning, a type of artificial intelligence (AI) learning method.

Since October 2015, Dentsu has collaborated with Data Artist Inc. to research and develop a system that predicts pre-broadcast TV ratings. This involved building deep learning models using training data such as historical ratings data, program genres, cast information, and online content viewing trends. The company has now successfully built a program that calculates predicted ratings within a high-speed data analysis platform. This enables stable prediction of ratings.

Moving forward, we will utilize "SHAREST (β version)" to conduct a verification project achieving high-precision ratings prediction for broadcast slots within one week in the Kanto region. We plan to expand this sequentially to the Kansai and Chubu regions.

Furthermore, in recent years, it has become possible to acquire various marketing data related to television media. As one application of this data, AI can learn the correlation between past TV viewing and marketing data. This enables the prediction and automation of ratings for specific targets (such as by gender and age group), which was previously difficult to achieve. This allows for the optimal allocation of advertisements for products targeting different audiences to the best commercial positions. In this project, we will conduct implementation verification using predicted viewership ratings within our company. In the future, we will demonstrate advanced operations for assigning TV commercial materials and enhancing advertising effectiveness.

Furthermore, when allocating commercial materials from multiple brands to advertising slots based on predicted viewership, the current practice often involves manual viewership forecasting, which presents the challenge of enormous operational burden. We anticipate that automating viewership prediction through AI will significantly enhance operational efficiency.

Moving forward, we will continue to leverage AI across various media-related marketing data to contribute to creating new value for our client companies.

<Concept Diagram of "SHAREST (β version)">

End

---

Dentsu Inc. News Release
http://www.dentsu.co.jp/news/release/2017/0616-009318.html