众源轨迹大数据隐含的信息知识对数据智能服务和城市管理决策具有重要价值。但大量粗质轨迹在数据清洗环节被删除丢弃，导致隐含其中且附属于主行程的衍生行为信息丢失，并带来数据挖掘结果偏差、语义知识解译不准等问题。为减少轨迹清洗造成的信息丢失，亟需实现由粗质轨迹清洗丢弃向知识萃取的升级转变。为此，本项目针对众源粗质轨迹中的衍生行为语义知识萃取问题，以“事件检测→行为识别→语义解译”为研究主线，发展面向粗质轨迹数据的知识萃取方法，包括：1）集成一致性检测模型、领域知识和轨迹上下文，建立行为事件检测提取方法，自动检测衍生行为事件。2）耦合推理模型和随机森林模型，提出行为事件分类识别方法，准确分类识别衍生行为事件。3）通过轨迹表征建模活动场景，发展时空场景与领域知识嵌入的行为场景解译方法，精准解译衍生行为语义。本课题有望突破粗质轨迹数据挖掘的技术瓶颈，为轨迹大数据分析提供新的研究视角。

The implied information and knowledge of crowdsourcing trajectory big data play an important role in data intelligent service and urban management decision. However, in the data cleaning process, a large number of raw trajectory data are discarded, resulting in loss of derived behavior information. It is also brings some problems, such as mining results bias, semantic interpretation inaccuracy and so on. In order to reduce the information loss which caused by track data cleaning, it is necessary to extract knowledge from raw trajectory data. Therefore, this project aims at the knowledge extraction with raw trajectory data. Taking the “event detection→behavior recognition→semantic interpretation” as the main research line, the knowledge extraction method for raw track data is proposed. Firstly, by integrating track feature consistency detection model, domain knowledge and track context, the detection and extraction method of behavior events is established to automatically detect the derived behavior event. Secondly, after coupling reasoning model and random forest model, the classification and recognition method of behavior events is proposed to accurately classify and recognize the derived behavior event. Thirdly, modeling activity scene based on trajectory representation, the behavioral scene interpretation method of embedding the domain knowledge and spatio-temporal scene is developed to accurately interpret the derived behavioral semantic. This project is expected to break through the technical bottleneck of raw track data mining and provide a new research perspective for track big data analysis.