融合图像和时空信息的社交媒体用户活动分类方法

社交媒体签到数据中蕴含着大量的用户活动信息。理解社交媒体用户的活动和行为类型，对探索人类的移动性和行为模式等有着重要意义。提出了一种针对新浪微博（简称为微博）的用户活动分类方法，结合图像表达和时空数据分类技术，识别微博签到数据所代表的用户活动类型。首先，根据兴趣点属性信息将微博签到数据所代表的用户活动分为餐饮、生活服务、校园、户外、娱乐、出行6大类；然后，基于卷积神经网络和K近邻分类方法，融合签到数据中的图像场景信息与时空信息，对微博用户的活动行为进行分类。实验结果表明，所提方法能够显著提高微博用户活动类型识别的准确性，为精确探索人类行为活动提供更加有效的数据支持。

Social Media User's Activity Classification Integrating Image and Spatiotemporal Information

  Objectives  Social media check-in data contains a large number of social media users' activity records. It is important to understand social media users' activities for exploring human mobility and behavioral patterns.  Methods  We present a user activity classification method on the check-in data of Sina Weibo, which is a very popular Chinese social network service, referred to as Weibo. This method achieves the goal of recognizing users' activities by combining image expression and spatiotemporal data classification technology on the Weibo check-in data. Firstly, we divide Weibo user activities into six categories, including catering, social service, education, outdoor, entertainment, and travel related, according to the points of interest attribution of check-in. Secondly, by applying the convolutional neural network and K-nearest neighbor method, the scene information in images and spatiotemporal information in check-ins are merged to classify the activity behavior of Weibo users.  Results  Experimental results show that the proposed method significantly improves the accuracy of Weibo user activity classification.  Conclusions  Although the classification method cannot comprehensively improve the performance of all user activity types, it can better express the microblog user activity and provide more effective data support for the exploration of human behavior.