Scenic discover and representation based on social media

Nowadays, travelling has been a way of relaxing. But we always can´t get the fully and timely information about the objective attractions. Meanwhile, social media services and web sites, such as Flickr and YouTube, provide plenty of photos, most of which are taken during the journeys. And every photo describes a perspective of an attraction. It has become a hot topic about how to discover scenic from the scattered pictures and then provide users with panoramic display of the tourist attractions. In this paper, we presents a novel method to discover hot spots of a given area based on social media data, and show them extracted keywords and photos. First, we use the k-Means clustering algorithm to cluster geo-tagged photos, and use the event-detection method to identify scenic. Then, we propose a text denoising framework and extract the most representative keywords for each cluster by employing an improved TF-IDF approach. Finally, in order to extract a set of representative images, we cluster and sort the photos with keywords using the visual information content. We use the most representative keywords and photos to represent attractions. Experimental results show that proposed method is effective.