文章简介：利用社交网站Flickr上照片的geotag信息将这些照片聚类发现城市里的旅游景点，通过各照片的拍照时间得到用户访问某景点时的时间上下文和天气上下文（利用时间和public API of Wunderground），将访问景点的上下文进行排序得到popular的上下文作为景点的上下文。在给用户作推荐时，首先得到用户当前的上下文或者要访问景点的上下文，利用上下文匹配出一些景点，然后在这些景点里头根据user-based collaborative filtering方法进行推荐，user-based collaborative filtering中用户对景点的评分使用用户访问某景点的次数。
The architecture behind our approach is configured into various modular tasks to carry out different operations as depicted in Figure 1.
We find tourist locations using spatial proximity of photos and enrich the aggregated locations with semantic annotations using textual tags annotated to photos in combination with information provided by Web services. Profiles of locations are built to describe the contexts in which they have been visited. To derive temporal context, geotags and temporal tags annotated with photos are exploited, whereas to derive weather context, we query thirdparty weather Web services to retrieve weather conditions. Relationship between users and locations is drawn to model users’ travel preferences. Then, these users’ preferences are used to estimate the similarities among users. For making recommendations, first we filter the locations based on contextual constraints, and then rank the locations by personalized score. A measure is defined to identify similar users in previously visited cities and aggregate these users’ opinions to obtain personalized score for each location in a target city for the target user.
