TPR-TF: time-aware point of interest recommendation model based on tensor factorization

摘要

ublishercopyright>? 2019, Jilin University Press. All right reserved.? 2019, Jilin University Press. All right reserved. With the rapid growth of the location-based social networks, Point of Interest (POI) recommendation has become an important research topic in the field of data mining. Existing approaches for POI recommendation task do not reasonably utilize the time sensitivity of POI recommendations and have not taken full account of the user's behavior preferences at different time periods, causing the POI recommendation performance is poor. Firstly, this paper studies the POI recommendation problem of time sensitivity and proposes a time dynamic partition algorithm based on hierarchical clustering. Through partition the fine grain of time, the result of the experiment is more reasonable and effective than the previous experiments which partition time is evenly given by experience. Secondly, by combining the time-aware recommendation with the influence of the user's direct friendship and potential friendship, the paper expands the scope of user's social influence, and then further improves the POI recommendation performance. Lastly, using the method of randomly selecting POIs by the frequency distribution of check-ins, it improves the classic BPR method. Experimental results on the two datasets indicate that the TPR-TF model is superior to the current mainstream POI recommendation models, in terms of precision and recall.With the rapid growth of the location-based social networks, Point of Interest (POI) recommendation has become an important research topic in the field of data mining. Existing approaches for POI recommendation task do not reasonably utilize the time sensitivity of POI recommendations and have not taken full account of the user's behavior preferences at different time periods, causing the POI recommendation performance is poor. Firstly, this paper studies the POI recommendation problem of time sensitivity and proposes a time dynamic partition algorithm based on hierarchical clustering. Through partition the fine grain of time, the result of the experiment is more reasonable and effective than the previous experiments which partition time is evenly given by experience. Secondly, by combining the time-aware recommendation with the influence of the user's direct friendship and potential friendship, the paper expands the scope of user's social influence, and then further improves the POI recommendation performance. Lastly, using the method of randomly selecting POIs by the frequency distribution of check-ins, it improves the classic BPR method. Experimental results on the two datasets indicate that the TPR-TF model is superior to the current mainstream POI recommendation models, in terms of precision and recall.