移动场景中用户出行意图预测与地点推荐的智能方法研究

Human mobility is a basic research problem concerned by smart city, geography and sociology. Social media, GPS trajectory and contextual data bring new opportunities for human mobility research. However, current studies lack intelligent deep mining and fusion analysis on multiple mobility data, and ignore the essential factor, mobility intentions, in human mobility. To overcome these limitations, we propose a new research object "mobility scene", which has heterogeneous structure, complex connections and dynamic evolution, and three new scientific questions of how to model mobility scene, how to predict mobility intentions and how to recommend locations according to intentions. To develop deep learning and other intelligent methods, mining contextual big data, large-scale social media and GPS trajectories, we establish natural language processing methods to model text description, feature vector and similarity network for mobility scene, intelligent methods of revealing complex associations in mobility scene to predict mobility intentions, statistical methods for fusing mobility intentions and preferences to recommend locations. To propose new problems, establish new perspectives and develop new methods, we reexamine human mobility via newly quantitative analysis methods to get further understandings, and to improve intelligence, real-time accuracy of mobility activities prediction and location recommendation. The study has important values not only for theoretical and methodological researches in intelligent decision support, but also for applications of smart travel, geographical location information services in smart city.

移动出行是智慧城市、地理学、社会学的基础研究问题。丰富的社交媒体、移动轨迹、情境数据为移动出行研究带来新机遇。然而当前研究缺乏对多种出行数据的智能深入挖掘和融合分析，更忽略出行意图这一关键因素。针对这些局限，本项目提出结构异质、关联复杂、动态演化的“移动场景”这一新研究对象，以及如何对移动场景建模、预测出行意图、根据出行意图推荐地点3个新科学问题。发展深度学习等智能方法，挖掘海量情境数据和大规模移动轨迹，建立基于自然语言处理方法的场景文本描述、特征向量和网络模型，揭示场景复杂关联的出行意图智能化预测方法，融合出行意图和偏好的智能地点推荐方法。研究通过“提出新问题-确立新视角-发展新方法”，重新审视、量化分析和理解人类移动出行规律，提高出行预测和地点推荐的智能性、实时性和准确性，不仅在智能决策支持领域有重要理论方法价值，而且在城市智能出行、基于地理位置信息服务等智慧城市领域有重要应用价值。

本项目围绕移动场景中用户出行意图预测与地点推荐中的若干关键科学问题，遵循理论与应用相结合的研究思路，从信息科学与系统科学的角度，应用统计分析、智能计算、深度学习、复杂网络、图神经网络等国际领先的理论方法，考察移动场景中用户出行活动所遵循的规律，开展预测用户意图与推断出行地点的智能化推荐方法等研究，构建融合多源异构数据的移动场景建模、考虑多种情境因素的用户偏好与出行意图预测、集成出行意图与场景网络的地点推荐等方法，攻克移动场景中多源异构数据的挖掘与表示、移动场景与出行意图的复杂关联建模、出行意图与场景网络的智能集成推荐等难题。项目取得一系列创新性研究成果，显著提高了移动场景中推荐系统的准确性和多样性，在移动大数据场景下推荐系统领域具有重要的理论意义；同时，项目扩展应用成果在电子商务、基于位置的社交网络、信息推荐等领域具有重要实际应用价值。成果方面，在Knowledge-based systems，Information Processing & Management，Expert Systems with Applications，Soft Computing，World Wide Web Journal，Knowledge and Information Systems 等相关领域SCI检索的国外重要学术期刊发表论文14篇，在管理科学、现代情报、数据分析与知识发现等相关领域CSSCI检索的国内重要学术期刊发表论文4篇，在Hawaii International Conference on System Sciences (HICSS)、Pacific Asia Conference on Information Systems (PACIS)、International Conference on Computer Information and Big Data Applications (CIBDA)等国际会议发表英文论文7篇。