A news recommendation method based on feature fusion and user interests

There are many types of news categories and user behaviors, and most existing news recommendation systems are based on news text representation. Therefore, a text and image fusion news recommendation system is proposed, which combines the category, title, abstract, image, and user's long and short-term interests through attention mechanism for feature fusion representation. Text is processed using the Albert model, images are processed using the Swin Transformer model, and user interests are processed using the GRU model. Through experiments on real datasets, this method shows significant performance improvements compared to other recommendation algorithms.