Recently, the international leading AI conference AAAI 2026 has accepted three research papers authored by the Spatiotemporal Big Data and Intelligence Team at the School of Computer Science and Engineering (School of Cyberspace Security). These studies delve into trajectory similarity learning, next point-of-interest (POI) recommendation, and machine unlearning techniques. One paper, titled 'Region-Point Joint Representation for Effective Trajectory Similarity Learning', introduces a trajectory similarity learning approach named RePo, which leverages joint point-region representation. By integrating structural and visual features, this method effectively tackles the challenge of capturing fine-grained trajectory information, resulting in an average accuracy boost of 22.2% across three real-world trajectory datasets. Another paper, 'TOOL4POI: A Tool-Augmented LLM Framework for Next POI Recommendation', presents a tool-augmented large language model framework, Tool4POI. This framework facilitates open-set recommendation capabilities by invoking external tools, achieving an Acc@10 metric exceeding 40% in 'out-of-history' recommendation scenarios. Lastly, 'Beyond Superficial Forgetting: Thorough Unlearning through Knowledge Density Estimation and Block Re-insertion' proposes an innovative unlearning framework, KUnBR. This framework achieves comprehensive unlearning through knowledge density estimation and block re-insertion strategies, attaining state-of-the-art (SOTA) performance on multiple benchmarks while preserving the model's general capabilities.