基于事件语义增强的地理信息服务群推荐

With the development of communication technologies and social network, there are more and more interactions between users. To satisfy the social requirement of users, the recommender systems have been more and more facing group users. Also, due to the popularization of different types of mobile terminals, spatial information has come to be one of the important factors in recommender systems. As for the group recommendation of geographic information services, the difficulties exist in the identification of dynamic groups, different human mobility patterns under different circumstances, and the complex in calculating semantic similarity and spatial similarity of items. .The project will research the methods of dynamic group’s identification and semantic enrichment of group events; analyze the spatial pattern of human activities; and study the method of group recommendation considering semantic and spatial information. The research includes: .1) Formalization of group event. Group events are the clusters of certain group of people in limit extend of space and time duration. By defining the temporal scale, the spatial scale and the strength of connection, we will realize the formal representation of group events..2) Identification of dynamic group events from location-based social network. We propose a distance-based algorithm of detecting communities in spatially constrained social networks. The algorithm can identify communities that are both highly topologically connected and spatially clustered, and thus can find dynamic group events in real time..3) Semantic enrichment of group events with multiple features. Multi-dimension of features, such as spatial signature, temporal signature and theme signature, will be used to enrich the semantics of the event. Topic model will be used to get the probability model of the context of events..4) Group recommendation system based on spatial heterogonous information network. Through aggregating the spatial mobility patterns and users’ preference, the common spatial and item preference of users will be acquired. Spatial heterogonous information network will be constructed, and then semantic meta-path similarity and spatial similarity items can be calculated. Based on the similarity matrix and preference metrics, group recommendation can be implemented..By the study of this project, we will solve the scale effect of group events, build spatial patterns and distance models of human activities in different contexts, and leverage the semantics of the recommendation system. Finally, a group recommendation system of geographic information services which considers semantic and spatial information will be implemented.

群推荐是为群体用户推荐其感兴趣的项目。随着位置社交网络和群体活动对人们生活的渗入，群体活动对于空间服务的需求不断显现，推荐系统越来越多从面向个体用户转为面向群体用户，地理空间也逐渐成为推荐系统中不可忽略的重要因素。针对群体的地理信息推荐面临群体事件识别困难、人类活动空间模式差别、考虑空间和语义的相似性计算复杂等问题。本项目建立群体事件的形式化表达，研究从空间社交大数据中挖掘事件的方法，并利用多维特征和主题模型方法对事件进行语义增强，从而获取事件的语境；研究不同语境下的空间行为模式；利用空间异构信息网的方式组织数据，以异构信息网中元路径的语义相似性实现推荐系统的语义提升；提出群体行为模式和群体偏好的聚集策略，构建推荐模型。通过研究，建立适用于推荐系统的不同语境下的活动空间和距离模型，以POI推荐为例，实现位置社交网络下实时发现群体聚集事件并考虑人们行为习惯和兴趣偏好的地理信息的群推荐。

随着位置社交网络和群体活动对人们生活的渗入，群体活动对于空间服务的需求不断显现，地理空间逐渐成为推荐系统中不可忽略的重要因素。针对地理信息群推荐面临的人类活动空间模式差别、活动类型语境差异等问题，项目提出从位置大数据感知人类活动和空间多重语义的方法，并且从大数据中分析活动空间模式和空间、属性偏好，以实现考虑时空语义信息的地理信息推荐。. 项目定义群体事件为一组有相互作用关系的对象在特定时间和空间范围内的活动，研究事件的形式化表达，建立特征空间分层的矩阵分解模型，并采用张量模型，表达事件时空间特征及多时空要素相互作用，提出基于多重矩阵和张量高阶模型的事件多重语境表达方式。在人类活动空间模式方面，项目基于地理位置大数据，采用流向量相似性计算的方法研究人口流动模式，采用张量分解的方法研究城市人群出行的时空模式及影响因素，挖掘城市间和城市内部人类活动模式，以及特定主题的人类活动模式。研究人类活动的语义增强方法，充分利用人类活动的时空多维度特征，从地理位置语义和行为语义两个方面对事件进行语义增强，推导出人们的活动类型，从而获取群体行为偏好，为满足群体偏好和认知的地理信息推荐提供依据。最终将人类活动模式和偏好用于地理信息推荐，实现基于矩阵分解模型的个性化POI推荐、基于高阶图信息传播模型的旅游景点推荐和基于Word2Vec模型的旅游行程推荐，并实现基于项目聚类的群推荐方法。. 随着移动定位技术的成熟，地理信息推荐的需求日益增加，本项目的研究成果在移动互联网和地理信息推荐方面具有较好的应用前景。