News recommendation is a new challenge in the current age of information overload. Making personalized recommendations from the sources of condense textual information is not trivial. It requires the understanding of both the news article’s semantic meaning, and the user preferences via the user’s history records. However, many existing methods are not capable to address the requirement. In this paper, we propose our novel news recommendation model called CUPMAR, that not only is able to learn the user-profile’s preferences representation in multiple contexts, but also makes use of the multifaceted properties of news articles to provide personalized news recommendations. The main components of the CUPMAR model are the News Encoder (NE) and User-Profile Encoder (UE). The NE uses multiple properties of a news article with advanced neural network layers to derive news representation. The UE infers a user’s long-term and recent preference contexts via her reading history to derive a user representation, and finds the most relevant candidate news for her. We evaluate our CUPMAR model with extensive experiments on the popular MIND dataset and demonstrate the strong performance of our approach. Our source code is also available online for the reproducibility purpose.