Browsing by Author "Rahimi, Seyyed Mohammadreza"

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Activity-based and Behavior-based Location Recommendation in Location Based Social Networks
    (2014-01-31) Rahimi, Seyyed Mohammadreza; Wang, Xin
    Location-Based Social Networks (LBSNs) are social networks with functionalities that let users share their location information with other users. Location recommendation is the task of suggesting unvisited locations to the users. A good location recommender should make user-specific recommendations based on users’ preferences, geographical constraints and time. In this thesis we investigate the development of two novel location recommendation methods for Location-Based Social Networks (LBSNs), the Probabilistic Category-based Location Recommender (PCLR) and the Behavior-based Location Recommender (BLR). The PCLR method finds the temporal and spatial patterns of users’ activities in the form of temporal and spatial probability distributions. It then uses the patterns to select the right category of locations and recommend nearby locations of that type to the user. On the other hand, the BLR method first extracts user behaviors from their check-in history. It then utilizes a collaborative filtering technique to extract common behaviors and predict behavior of the user at a given time. Finally, BLR filters locations in the user’s proximity based on the predicted behavior when making the location recommendation. PCLR and BLR methods go through a set of experiments on a real-world check-in dataset. These experiments show that PCLR and BLR methods improve the performance of the existing location recommenders in terms of precision and recall. Additionally, the BLR method produces much better recommendations for the cold-start users.
  • Thumbnail Image
    ItemOpen Access
    Behavior-based and Contextual Location Recommendation for Location Based Social Networks
    (2021-02-10) Rahimi, Seyyed Mohammadreza; Wang, Xin; Far, Behrouz Homayoun; El-Sheimy, Naser; Liang, Steve H. L.; Uddin, Gias; Wachowicz, Mónica
    Location-Based Social Networks (LBSNs) are social networks with functionalities that let users share their location information with other users. A service capable of improving user engagement and bring more check-ins to the Location-based Social Network is location recommendation. Location recommendation is the task of suggesting unvisited locations to the users. An effective location recommendation model makes user-specific recommendations based on users’ preferences, geographical constraints and contextual information such as time and weather. The main question we need to answer to design a location recommendation model is how to effectively utilize different types of information into location recommendation. In this thesis, Behavior-based Location Recommendation (BLR) and Contextual Location Recommendation (CLR) are proposed, these models effectively utilize temporal and other contextual information to produce improved location recommendations. In this thesis we investigate the development of two novel location recommendation methods for Location-Based Social Networks (LBSNs), the Behavior-based Location Recommendation (BLR) and the Contextual Location Recommender (CLR). The BLR finds the temporal and spatial patterns of users’ behaviors in the form of temporal and spatial probability distributions. It then uses the patterns to predict the location type and recommends nearby locations of that type to the user. On the other hand, the CLR method first extracts the responses of the users to contextual triggers using their check-in history. It then utilizes a tensor factorization technique to extract common responses and predict the user response with the given set of contextual triggers. Finally, CLR filters locations in the user’s proximity based on the predicted location type. To find user similarities, both BLR and CLR utilize Random Walk with Restart. To improve the performance of these methods, an optimized random walk with restart method is also proposed that can improve the time complexity of random walk with restart by a factor of at least 6.75. Both BLR and CLR methods go through a set of experiments on a real-world check-in dataset. These experiments show that BLR and CLR methods improve the performance of the existing location recommendation methods in terms of precision and recall. Additionally, both BLR and CLR methods can achieve higher precision and recall values for cold-start users compared to the well-known baseline models.