News recommendation (NR) is critical for helping users to navigate the vast amount of information on online news platforms. However, the key challenges of tackling the cold-start problem, comprehensively modeling user interests and accurately encoding news semantics in NR have not yet been effectively addressed by existing methods. In this paper, we propose to alleviate the cold-start problem by exploiting news popularity. We also argue that the browsing history of a user can be organized as a personalized hypergraph to augment user interest modeling. Furthermore, we suggest enriching the representation of a candidate news article by retrieving its similar news. Our proposed NR method, named P opularity P rediction with S emantic R etrieval (PPSR), features a retrieval-driven popularity predictor that leverages click records to predict news popularity by incorporating semantic retrieval. It also comprises a hypergraph-based user encoder that mines the rich and diverse relatedness among news in a user’s browsing history. In addition, it designs a semantic-enhanced news encoder that retrieves and utilizes similar news in the training set to enrich semantic encoding for candidate news. Experiments on real-world datasets show that our PPSR can outperform the state-of-the-art methods in terms of higher recommendation accuracy and can well mitigate both the user-level and news-level cold-start problems.

Recent advances in implicit neural representations have achieved impressive results by sampling and fusing individual points along sampling rays in the sampling space. However, due to the explosively growing sampling space, finely representing and synthesizing detailed textures remains a challenge for unbounded large-scale outdoor scenes. To alleviate the dilemma of using individual points to perceive the entire colossal space, we explore learning the surface distribution of the scene to provide structural priors and reduce the samplable space and propose a Point Diffusion implicit Function, PDF, for large-scale scene neural representation. The core of our method is a large-scale point cloud super-resolution diffusion module that enhances the sparse point cloud reconstructed from several training images into a dense point cloud as an explicit prior. Then in the rendering stage, only sampling points with prior points within the sampling radius are retained. That is, the sampling space is reduced from the unbounded space to the scene surface. Meanwhile, to fill in the background of the scene that cannot be provided by point clouds, the region sampling based on Mip-NeRF 360 is employed to model the background representation. Expensive experiments have demonstrated the effectiveness of our method for large-scale scene novel view synthesis, which outperforms relevant state-of-the-art baselines.