Popularity prediction of video content and clustering using networks

Abstract

The content market, including video content market, is a high-risk, high-return industry. Because the cost of copying and distributing the created video content is very low, large profit can be generated upon success. However, as content is an experience good, its quality cannot be judged before purchase. Hence, marketing has an important role in the content market because of the asymmetry of information between suppliers and consumers. Additionally, it has the characteristics of One Source Multi Use; if it is successful, additional profits can be created through various channels. Therefore, it is important for the content industry to correctly distinguish content with a high probability of success from the one without it and to conduct effective marketing activities to familiarize consumers with the product. Herein, we propose a methodology to assist in data-based decision-making using machine learning models and help in identifying problematic issues in video content markets such as movies, TV programs, and over-the-top (OTT) market. In the film market, although marketing is very important, decisions are still made based on the sense of practitioners. We used the market research data collected through online and offline surveys to learn a model that can predict the number of audiences on the opening-week Saturday, and then use the learned model to propose a method for effective marketing activities. In the TV program market, programming is performed to improve the overall viewership by matching TV programs and viewer groups well. We learn a model that predicts the audience rating of a program using the characteristics of the program and the audience-rating information of the programs before, after, and at the same time, and use the resulting data to assist in decision-making to find the optimal programming scenario. The OTT market is facing a new problem of user's perception bias caused by the recent recommendation system. In the fields of politics and news particularly, if the user does not have access to different viewpoints because of the recommendation service, it may create and/or deepen a bias toward a specific political view without the user being aware of it. In order to compensate for this, it is important to use the recommended channel while the user is well aware of what kind of channel it is. We built a channel network in the news/political field using the data extracted from the comments left by users on the videos of each channel. In addition, we propose a method to compensate for the bias by classifying networks into conservative and progressive channel clusters and presenting the topography of the political tendencies of YouTube channels.