Balanced news neural network for a news recommender system

News recommender systems face unique challenges due to the rapidly changing readers' interests over time. Some of the reader's interests are long-term, and some are short-term that need to be addressed in a news recommender system. Diversification is also required in a news recommender system to keep readers engaged in the reading process and expose them to various viewpoints. We propose a deep neural network for the news recommendation problem that learns multi-faceted news representations from the news content. The proposed model also learns the reader's long-term interests from the whole click history and the short-term ones from the click history using LSTMs. The attention mechanism is used to learn a reader's diversified interests. We give different levels of attention to the news and reader components. Experiments on two news datasets have shown the superiority of our proposed method compared to state-of-the-art methods.