Time-aware Point-of-interest Recommendation

Quan Yuan, Gao Cong, Zongyang Ma, Aixin Sun, Nadia Magnenat-thalmann
Proceeding of the 36th International ACM SIGIR Conference

Introduced

With the increasing popularity of mobile devices and the large number of geographically based social applications being used by users, people have left a lot of mobile data on the Internet, especially the user's sign-in information. Based on these registration information, Internet companies can provide consumers and businesses with high-quality location-based personalized services, such as providing consumers with personalized referral services to help them find new and interesting entertainment venues, while providing businesses with potential consumers, customized personalized ads.

The main innovation points of this article include the following three points:

• A new research problem is defined, that is, given a target user and given a time, the system needs to recommend a new POI point of interest based on the time context for the target user at that time;
• Through the historical check-in data, this paper analyzes the time and space influence of the crowd registration behavior, and puts forward a POI recommendation framework which integrates the influence of space and time.
• Through experiments in two datasets in Foursquare and Gowalla, the article proves:
  1. The time factor is helpful to improve the accuracy of POI recommendation algorithm;
  2. The algorithm in this paper is better than other algorithms considering the space influence factors.
  3. The proposed framework, which considers both time and space factors, maximizes the accuracy of the algorithm.

Algorithm process symbol definition

symbols	Description
U,l,t	User set, POI set, time slot set
U,v,l,t	User \ (U,v\in u\), POI \ (l \in l\), time slot \ (T \in t\)
\ (c_u\), \ (c_{u,t}\)	The binary check-in vector of U over L, and the Binaray check-in vector u over l at t
\ (c_{u,l}\), \ (c_{u,t,l}\)	element of \ (c_u\) and \ (c_{u,t}\), respectively
\ (w_{u,v}\)	The similarity between U and V
\ (w^{(t)}_{u,v}\), \ (w^{(TS)}_{u,v}\)	time-enchanced Similarity, smoothed similarity
\ (dis (l_i,l_j) \)	Distance between \ (l_i\) and \ (l_j\)
WI (DIS)	The willingness a user visits a dis far away POI
\ (ci_l\), \ (ci_{l,t}\)	The set of Check-ins at L, \ (ci_l\) at time t

Time-aware Point-of-interest Recommendation