Details

Autor(en) / Beteiligte

• Chen, Jiangmei
• Zhang, Wende

Titel

TeSP-TMF: A temporal-aware personalized POI recommendation approach based on potential preferences and grey relational analysis

Ist Teil von

• Electronic commerce research and applications, 2023-03, Vol.58, p.101243, Article 101243

Ort / Verlag

Elsevier B.V

Erscheinungsjahr

2023

Link zum Volltext

Quelle

Access via ScienceDirect (Elsevier)

Beschreibungen/Notizen

• •A temporal-aware personalized approach is proposed for accurate recommendations.•Users’ potential preferences are investigated to relieve data sparsity.•Grey relational analysis is leveraged as a similarity evaluation under sparse datasets.•A new approach combined grey relational analysis and matrix factorization is developed.•Experimental results on two real-life datasets verify the superiority of our method. The great advances in mobile devices make it critical to exploit multiple auxiliary information for efficient point-of-interest (POI) recommendations. However, most existing methods can neither effectively address the extreme data sparsity encountered in temporal-aware recommendation nor reasonably deal with implicit feedback. Furthermore, previous methods, including those relying on similarity measurement, cannot capture the intrinsic data correlations exhibited in the sparse data. To remedy these drawbacks, this paper proposes a temporal-aware personalized recommendation framework called Temporal Spatial Popularity and Temporal-aware Matrix Factorization (TeSP-TMF). In TeSP-TMF, we present the potential preferences model and temporal-aware matrix factorization model. Specifically, users’ potential preferences that incorporate temporal, spatial, and popularity information are investigated to learn a set of potential POIs for each user, which are filled into the original matrix for further matrix factorization, owing to the sparse nature of the rating matrix. On this basis, we develop a new approach using a combination of grey relational analysis and matrix factorization, which not only relieves data scarcity caused by dividing the matrix with time slots but also strategically mines inherent temporal relationships. Finally, we implement extensive experiments on two public datasets collected from Foursquare and Yelp. Experimental results demonstrate that our proposed method yields competitive quality compared to several state-of-the-art methods. Theoretically, our work contributes promising insights for mitigating data sparsity. In practice, our results can enhance personalized temporal-aware POI recommendation services for location-based social networks.