基于轨迹数据的空间移动行为预测与位置推荐

Trajectory data were widely employed in behavioral geography, time geography, even in psychology to research human mobility behaviors. In this study, we used trajectory data and put forward a new idea that was spatial mobility behavior prediction and personalized location recommendation by the means of trajectory mining. We improved the methods of trajectory modeling and semantization by blending land use, POI/ROI and other map information, calculated maximum revisited probability on the historical position by mining the individual space movement mode and realized location prediction. And then, we computed the users’ similarity by trajectory similarity and location similarity which was inferred indirectly by the visited times of similar users, subsequently achieved the positional recommend through collaborative filtering algorithm. The context and transposition patterns were the important factors to adjust the selection sequence to location recommended , so we combine them to reconstruction the chain of multiply spatial position and give the personal visual representation. To evaluate the effectiveness of location prediction and recommendation, a customer satisfaction indicator was introduced into the quality assessment family to cooperate with the accuracy, coverage, freshness and accidental indicators to accomplish the adaptive feedback of the user model and the recommendation algorithm. This study aims to establish the theory and method system of the spatial information recommendation, promote the flexibility of the spatial information service and realize the transformation from “I am looking for information” to “The information is looking for me”.

课题通过轨迹数据挖掘进行空间移动行为的预测并藉此实现位置的推荐。主要研究内容包括：构造轨迹建模与语义化计算的方法，通过挖掘个体的空间移动模式计算历史位置的最大重访概率从而实现位置的预测。依据轨迹相似性解算的空间相似性和社交网络解算的关系相似性拟定出用户的相似性，并结合相似用户对候选位置的兴趣度，采用协同过滤实现位置推荐。顾及上下文环境和出行模式，聚合位置预测与推荐的结果，构造时间序列上的动态多重空间位置链，实现个性化的表达。为评价位置预测与推荐的精度，结合准确度、覆盖率、新鲜性和意外性等评价指标，引入预测与推荐位置上的驻留时间及其它因子，通过结构方程方法拟合出“用户满意度”这一综合指标以表征预测与推荐的效果，并据此实现预测与推荐的动态反馈。课题研究不仅为当前大数据背景下的人类移动行为分析提供理论与方法支撑，更是位置服务模式的重大变革，旨在促进位置服务从“我找信息”到“信息找我”的转变。

人群移动模式的探测与分析一直是交通出行、城市规划等领域的关键问题，也是地理学长期关注的人地关系研究的重要命题，大数据的出现给这些问题的研究提供了崭新的研究思路和数据支撑，轨迹数据作为典型的位置大数据形式成为研究上述问题的重要入口。课题主要针对如下内容开展了研究：1）发展了轨迹数据的语义化建模方法，提出了一种R-HDBSCAN聚类算法，解决了人群驻留点的合理提取问题；2）提出了基于轨迹数据探索人群空间移动模式的技术框架，进行了不同地理尺度下旅游人群的空间移动模式分析；3）提出了Fuzzy-LSTM Network方法，实现了人群空间移动位置的预测；4）基于标签进行了用户相似性的计算并通过用户相似性进行了位置推荐，基于特定用户的历史出行轨迹发展了神经网络方法实现了下一个位置的推荐，比较了不同推荐算法的精度；5）构建了用户情感计算模型，分析了用户的满意度。课题研究不仅为深度把控人群移动规律提供了方法支持和应用案例，也为城市的智慧化管理提供了新的决策依据。