This paper contains development of methods of improving TV-watching experience using Machine Learning for Linear TV recommendations. TV is a shared device, so the recommendations based on taste of only one user, may not be sufficiently effective. This paper discusses how Data Mining viewing data to improve the effectiveness of the advisory system. The main idea is to get new insights about the users, thus creating new knowledge that can be used for more accurate recommendations. This paper describes the methods of data mining on watching TV subscribers, as well as measures the effectiveness of the use of derived hypotheses in existing approaches to building recommender systems. Usage of such recommender could completely transform the TV advertisement and creates a new synergy between e-Business and TV by defining who is in front of TV and what he likes. From the other hand Smart TV allows user to make an order right from his TV-set that dramatically improves conversion rates. Existing methods and new method effectiveness are compared with offered approach by analyzing real people content consumption during one year.