大規模ユーザの滞在情報に基づくエリアの特徴付けとCOVID-19による影響分析

都市や地域の各エリアに人がどのように滞在し，どのように利用しているかを知ることは，マーケティングや商圏分析，出店判断などの商業的利用に加え，都市設計や交通政策などの自治体の政策判断など，幅広い分野で有用であり，これまでは大規模なアンケート調査などで実現されてきた．しかし，時間経過にともなう様々な事象（都市の発展，季節の変化，パンデミックの発生）によるエリア利用形態の変化を検知するには多大なコストが必要となり，その定常的な利用は現実的ではない．本研究は，大規模ユーザの移動履歴情報から滞在情報を抽出し，エリアを特徴付ける手法（Area2Vec）を提案する．本手法はWord2Vecアルゴリズムを参考として，到着時刻と滞在時間の情報からエリアの分散表現を構築する．この分散表現により，都市の機能分類に加え，滞在行動の時間的変化の観測・分析が可能になる．評価実験として，宮城県仙台市の国分町を対象にしたエリアの機能分類を行い，その有用性を確認した．また，COVID-19による人々の滞在行動の変化を調査するため，国分町と東京都新宿区の歌舞伎町を対象に検証を行い，人の滞在数の減少に加え，滞在時刻の夜から昼へのシフトを確認した．

Understading how people stay and use each area in a city or region is not only for commercial use such as marketing, trade area analysis, and store opening decision, but also for local governments such as urban design and transportation policy. It is useful in a wide range of fields such as policy making, and has been realized by large-scale questionnaire surveys. However, it requires a great deal of cost to detect changes in area usage patterns due to various events over time (urban development, seasonal changes, pandemics), and its regular use is not realistic. This study proposes a method (Area2Vec) that extracts stay information from the movement history information of large-scale users and characterizes the area. This method uses the Word2Vec algorithm as a reference to construct a distributed representation of the area from the arrival time and stay time information. This representation makes it possible to observe and analyze temporal changes in staying behavior in addition to functional classification of cities. As an evaluation experiment, we classified the functions of the area in Kokubuncho, Sendai City, Miyagi Prefecture, and confirmed its usefulness. In addition, in order to investigate changes in people's staying behavior due to COVID-19, we conducted a verification targeting Kokubuncho and Kabukicho in Shinjuku-ku, Tokyo, and in addition to reducing the number of people staying, from night to noon at the time of stay. The shift of was confirmed.