Personalized news recommendation device and method based on news content and theme feature

The invention discloses a personalized news recommendation device and method based on news content and theme feature. The recommendation device is equipped with seven modules, namely a news capturing module, a pre-treatment module, a theme model training module, a theme model predicting module, a user model building module, a news recommendation module and a recommendation treatment module. The recommendation method comprises the following steps: building an personalized user model with the theme model and a relevant named entity noun sequence to express the interest preference of the user reading news, and calculating weight and converting the theme feature vector of users so as to reduce the influence of hot themes and single news content on the user interest, thereby effectively overcoming the defects of concentrated user interest and insufficient diversity of recommendation results. In a recommendation output stage, an initial recommendation news list is treated, a theme grouping process based on the personalized user model is added on the basis of currently repeating data deleting and redundancy filtering, and news texts are reordered again according to the aging weight so as to recommend the accurate, diversified and novel personalized news.