News Session Recommendation Based on Long and Short-Term User Interest Representation and Multi-View Learning

In the current context of proliferation of news information, personalised news recommendation is important to help users quickly access content of interest. Existing approaches tend to ignore the diversity of users' reading behaviours and the variability of news topics within a session. They tend to assume that all the news read or clicked within a session are attributed to one main interest, while in reality, each news session may cover a variety of news with different interests corresponding to different topics. The paper introduces a news session recommendation approach that utilizes both long-and short-term representations of user interests, incorporating multi-view learning techniques, and the method treats the clicking behaviours of anonymous users as a session: initially, a multi-scale convolutional neural network is employed to capture the features of user-clicked news articles from mul-tiple viewpoints; Subsequently, a GRU multi-channel interest decoupler is crafted to identify potential interests within each clicked news item, enabling the modeling of diverse interests to capture the user's long-term preferences. Concurrently, recent user-clicked news items are considered as short-term interests. These long and short-term interests are then amalgamated to create the ultimate user interest representation; Finally, the user's preference value for the candidate news is calculated by click-summing and the final recommendation result is generated. By comparing with the benchmark method on the real data set, the findings demonstrate strong performance on objective evaluation metrics such as Area Under the ROC Curve (AUC) and Normalised Discounted Cumulative Gain (NDCG), affirming the efficacy of the proposed method for news recommendation.