Article
Graph-Based Personalized News Recommendation Using Deep Contextual Embeddings
With the overwhelming amount of digital news available today, helping users discover relevant content has become increasingly important. This work proposes a personalized news recommendation system that combines contextual text understanding with graph-based learning. The model leverages the Microsoft MIND dataset to capture user reading patterns, including both interacted and ignored articles, along with rich news attributes such as category and textual content. To better understand the semantics of news articles, a pre-trained BERT model is used to generate meaningful text representations. These representations are then structured into a graph, where relationships between users and news items are modelled and refined through a Graph Convolutional Neural Network . By learning from these complex connections, the system is able to capture deeper patterns in user preferences. Compared to conventional approaches like LSTM and CNN, the proposed method demonstrates improved recommendation quality by effectively integrating relational and contextual information. The results highlight the model’s ability to deliver more accurate and personalized news suggestions, making it a promising solution for modern content recommendation systems.
Full Text Attachment
