Modeling Trans-Device Content Experience and Knowledge Development for Detection of TV Audiences

Household audience ratings are the currency of the television broadcasting business. Information communication technology has rapidly changed television viewing since the beginning of this century, and the reproducibility of audience ratings as media indicators is beginning to be questioned. This is because television viewing through various devices must be considered, due to the decline of simultaneous viewing at a fixed time. This research applies artificial intelligence to explore several questions. What viewing features span media and devices? What are the viewer attitude models inherent in those features? What are the functions and rules driving the model? Can these be converted into knowledge, and can new hypotheses be obtained from this knowledge? For this, the technologies of natural language processing, neural network, and automated reasoning are applied. From data sequences including property amounts representing viewing attitudes, we visualize time-series feature patterns and extract a state space model driving these features and logical rules between the variables. This research is an example of the real-world application of a technical framework that combines symbolic inference and feature extraction, opening possibilities for more advanced reasoning.