An Expectation-Maximization approach for matrix factorization in location-based recommender systems
Year: 2022
Type: preprint
Abstract: On Location Based Social Network (LBSN) platforms, users benefit from location recommendation algorithms to find places of interest. Several algorithms have been proposed to accomplish the location recommendation task. However, these algorithms suffer from (i) the lack of trust information consideration and (ii) assuming users’ interest are static and does not change overtime. To overcome these two shortcomings this paper introduces an adaptive probabilistic-based location recommendation algorithm known as EXMAX which uses expectation-maximisation along with matrix factorisation to predict the rating a user may give to the next visit. The predicted rating will be used as a filter to remove any places with a score lower than the user’s predicted rating. EXMAX represents check-in data as a user-location matrix where every time a user checks into a new location check-in along with location information will be appended to the user-location matrix. This representation method ensures EXMAX preserves the most up-to-date users’ preference information. Also, the proposed algorithm exploits friendship network data as well as sentiment of the comments left by the user to label every checkin with either a positive or negative experience. To measure EXMAX efficacy, its performance is tested in several experimentation scenarios and compared against three rival algorithms in terms of root mean square error and Yelp dataset. The results confirm EXMAX robustness and superiority over other probabilistic and statistic location recommendation algorithms.
Source: Research Square
Author: Alireza Moayedikia
Institution: Swinburne University of Technology
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4310188914