A study on automatic TV scheduler personalization based on HMM for intelligent broadcasting services

Abstract

Television broadcasting services are entering a new era where technology makes a TV viewer being overwhelmed by floods of information. This situation is not easy to handle and is not welcomed. A particular service that can make the life of TV consumers easy by gathering information and adaptively proposed a selection of content for appropriate individual is needed. This thesis attempts to make a contribution in an ever increasing research to answer that necessity by proposing a method based on Hidden Markov Model (HMM). Markov model in general is well known and being used to model and predict processes based on available past behavior or pattern information. For eight weeks in advance, information such as time, duration, channels and content genres that the person watched are recorded. Based on that information we make a model to predict sequence of genres the person will watch in the following week. While the viewer is watching television, it would be apparent that our predictions may not be accurate. Based on selection of contents that contradicts our prediction which the viewer made, we can refine the HMM and generate new predictions. To asses the method, experiment was conducted using data gathered from selections of people. The resulting predictions were compared to the list of content``s information which the viewer watched. The method shows a satisfactory prediction result for viewers who showed consistent behavior or preference. The adaptive method showed a comparable performance and may increase predictions accuracy of the less consistent viewers.