In order to recommend TV programs and advertisements for TV audiences more efficiently, a new method of constructing diversified and accurate TV user profiles is proposed. In our
method, the micro-blog users and the TV users will be treated as the same users because they are concerned about the TV programs; hence, the tags of micro-blog users are obtained by the
web crawler first, then micro-blog users data are applied to build the model and use this model to predict the TV user's tags. In order to evaluate the accuracy of user profiles, we use the real viewing logs for a month. Our method is then evaluated by the content-based recommendation system. Experimental results show, compared with other algorithms, our method features better performance in the Precision and Area Under Curve (AUC). Thus, the method of using micro-blog data to construct TV user profiles is an effective solution.