TY - GEN
T1 - Point of interest recommendation by exploiting geographical weighted center and categorical preference
AU - Mo, Fan
AU - Yamana, Hayato
PY - 2019/11
Y1 - 2019/11
N2 - Point of interest (POI) recommendation is one of the indispensable services in location-based social networks (LBSNs). POI recommendation helps users find new locations and better understand the city. In LBSNs, the aspects, such as geographical information and categorical information, improve the accuracy of POI recommendation. In this paper, we propose two new techniques to improve the recommendation accuracy; 1) weighted center of a target user's each active area and 2) category-dependent threshold for categorical preference. The weighted center represents density-based center of a target user's active area. The geographical aspect usually adopts the target user's active areas that he frequently visited. Although previous researches define the active area by its active center and its radius, they choose the location of the most frequently visited POI as the active center even if there exist several POIs that have similar number of check-ins, which results in miss-definition of active center. Our weighted center is able to handle the target user's check-in probability, which follows a power-law distribution. Besides, previous researches predict users' preference for categories; however, they neglect the fact that different categories have different users' preference distributions. For example, a specific category has wide-range of subcategories to be preferred by user, but another category has a few subcategories to be preferred, even if there are many subcategories in the category. Thus, we set different thresholds to select candidate subcategories in each category. Experimental result with Weeplaces dataset shows that our method outperforms other baselines by at least 16.93% in F1-score@5.
AB - Point of interest (POI) recommendation is one of the indispensable services in location-based social networks (LBSNs). POI recommendation helps users find new locations and better understand the city. In LBSNs, the aspects, such as geographical information and categorical information, improve the accuracy of POI recommendation. In this paper, we propose two new techniques to improve the recommendation accuracy; 1) weighted center of a target user's each active area and 2) category-dependent threshold for categorical preference. The weighted center represents density-based center of a target user's active area. The geographical aspect usually adopts the target user's active areas that he frequently visited. Although previous researches define the active area by its active center and its radius, they choose the location of the most frequently visited POI as the active center even if there exist several POIs that have similar number of check-ins, which results in miss-definition of active center. Our weighted center is able to handle the target user's check-in probability, which follows a power-law distribution. Besides, previous researches predict users' preference for categories; however, they neglect the fact that different categories have different users' preference distributions. For example, a specific category has wide-range of subcategories to be preferred by user, but another category has a few subcategories to be preferred, even if there are many subcategories in the category. Thus, we set different thresholds to select candidate subcategories in each category. Experimental result with Weeplaces dataset shows that our method outperforms other baselines by at least 16.93% in F1-score@5.
KW - Categorical aspect
KW - DBSCAN
KW - Geographical aspect
KW - Location-based social networks
KW - POI recommendation
UR - http://www.scopus.com/inward/record.url?scp=85078727275&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85078727275&partnerID=8YFLogxK
U2 - 10.1109/ICDMW.2019.00021
DO - 10.1109/ICDMW.2019.00021
M3 - Conference contribution
AN - SCOPUS:85078727275
T3 - IEEE International Conference on Data Mining Workshops, ICDMW
SP - 73
EP - 76
BT - Proceedings - 19th IEEE International Conference on Data Mining Workshops, ICDMW 2019
A2 - Papapetrou, Panagiotis
A2 - Cheng, Xueqi
A2 - He, Qing
PB - IEEE Computer Society
T2 - 19th IEEE International Conference on Data Mining Workshops, ICDMW 2019
Y2 - 8 November 2019 through 11 November 2019
ER -