Context-Aware Location Recommendation Using Geotagged Photos in Social Media

Abstract

Abstract:Recently, the increasing availability of digital cameras and the rapid advances in social media have led to the accumulation of a large number of geotagged photos, which may reflect people's travel experiences in different cities and can be used to generate location recommendations for tourists. Research on this aspect mainly focused on providing personalized recommendations matching a tourist's travel preferences, while ignoring the context of the visit (e.g., weather, season and time of the day) that potentially influences his/her travel behavior. This article explores context-aware methods to provide location recommendations matching a tourist's travel preferences and visiting context. Specifically, we apply clustering methods to detect touristic locations and extract travel histories from geotagged photos on Flickr. We then propose a novel context similarity measure to quantify the similarity between any two contexts and develop three context-aware collaborative filtering methods, i.e., contextual pre-filtering, post-filtering and modeling. With these methods, location recommendations like "in similar contexts, other tourists similar to you often visited . . . " can be provided to the current user. Results of the evaluation with a publicly-available Flickr photo collection show that these methods are able to provide a tourist with location recommendations matching his/her travel preferences and visiting context. More importantly, compared to other state-of-the-art methods, the proposed methods, which employ the introduced context similarity measure, can provide tourists with significantly better recommendations. While Flickr data have been used in this study, these context-aware collaborative filtering (CaCF) methods can also be extended for other kinds of travel histories, such as GPS trajectories and Foursquare check-ins, to provide context-aware recommendations.

A Clustering-Based Context-Aware Recommender Systems Through Extraction of Latent Preferences

“…Their basic recommendation framework is independent of the clustering method, but they used a context-dependent variant of hierarchical agglomerative clustering which takes into account the user's current navigation context in cluster selection. Huang [13] explored context-aware methods to provide location recommendations matching a tourist's travel preferences and visiting context. The author specifically applied clustering methods to detect touristic locations and extract travel histories from geo-tagged photos on Flickr and then proposed a novel context similarity measure to quantify the similarity between any two contexts and develop three context-aware collaborative filtering methods, i.e., contextual pre-filtering, post-filtering and modeling.…”

“…The two error metrics is defined in (13) and (14) where the predicted rating, p i u, , for user u on item i is subtracted from the actual rating, r i u, ,as contained in the test set over the total number of ratings N on the item set. We apply min-max technique to scale the predicted rating between 0 and 1 so that we can be able to compare the evaluated approaches directly.…”

Recommending Locations Based on Users’ Periodic Behaviors

“…In Example 2, we introduce an example for explaining the progress of Algorithm 2. (6) if [ ] == ts do (7) map.put ( , ) (16) for t2 in T do (17) sp = t2-t1//suspected periods (18) if sp > 0 and sp < len(S)/2 do shown in the following based on the algorithm shown in Algorithm 2: (3)…”

“…So the algorithm will not judge the same periodic behaviors. The hitting items are M[3] [7], M[7] [11], and M[11] [15]. After this judging, the algorithm will judge the next item by the same method.…”

“…Zhang et al [6] proposed a personalized and efficient geographical location recommendation framework called iGeoRec. Huang [7] proposed a novel context similarity measure to quantify the similarity between any two contexts and develop three context-aware collaborative filtering methods for recommending locations. Yuan and Li [8] proposed a location recommendation algorithm based on temporal and geographical similarity in location-based social networks.…”