地理空間情報分野のトップカンファレンスであるACM SIGSPATIA2025で、個々人の将来の移動を予測するタスクがGISCUP 2025として開催されました。 提案手法では、Masked Language Modelを位置情報予測に応用したLP-BERTをベースとします。 LP-BERTではメッシュに区切られた座標を予測するため、Cross Entropyを損失関数としていますが、提案手法ではマクロな人口分布を制約として追加することで予測精度の向上を図りました。
2025
-
Trajectory Prediction Using Spatiotemporal BERT Leveraging Collective Trajectories
Keiichi Ochiai
In Proceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, The Graduate Hotel Minneapolis, Minneapolis, MN, USA, 2025
Human mobility prediction is a crucial research topic with wide-ranging applications, from urban planning to infectious disease forecasting. While the proliferation of GPS-enabled devices has made large-scale trajectory data available, a lack of standardized open-source datasets has hindered fair comparisons. Recent winning solutions in Human Mobility Prediction Challenges have often leveraged powerful models like BERT and GPT, but these approaches primarily focus on predicting individual movements, often neglecting the broader context of collective human flow.To address this limitation, we introduce an approach that leverages aggregated trajectories as a constraint to enhance individual trajectory prediction. Our solution is comprised of two key components: (1) a loss function that incorporates the geographical distribution of the aggregated trajectories, and (2) two additional features — the coordinates of the most populous location and a flag indicating the final record of the day. we confirmed that each of these components is effective in improving the GEO-BLEU score. Our final solution ranked 10th-place in the GISCUP competition 2025.
