The field of personalized recommendation systems is rapidly evolving, with a focus on improving user experience and engagement. Recent developments have seen a shift towards more fine-grained and nuanced modeling of user behavior, including the use of multimodal data and hyperbolic space to capture complex user preferences. Additionally, there is a growing interest in serendipitous recommendation systems that balance usefulness and unexpectedness to provide users with novel and engaging content. Notable papers in this area include the proposal of a novel approach to estimate user curiosity and provide serendipitous recommendations, as well as the development of a behavior-bind quantization method for multi-modal sequential recommendation. The HyperLLM model, which integrates large language models with hyperbolic space, has also shown promising results in capturing hierarchical information and improving recommendation performance. Furthermore, the GTS-LUM model has demonstrated effectiveness in modeling long-term and periodic user behavior sequences in the telecommunications industry. Overall, these advances have the potential to significantly enhance the accuracy and diversity of personalized recommendations, leading to improved user satisfaction and engagement.