Point-Of-Interest Recommendations: Learning Potential Check-Ins From Networks

The emergence of Location-based Social Network (LBSN) services presents a great opportunity to develop personalized Point-of-Interest (POI) recommender systems. Although a personalized POI recommender system can significantly improve users' experience for outdoor activities, it faces numerous challenges, including the availability of real POIs, and the difficulty of addressing user-location cold-start issues. To address these issues, we propose a two-step framework for leveraging POI information into a recommender system. First, we propose learning a set of real locations where an individual has previously checked in. The goal is to detect real and fake POIs and feed the factual information into the subsequent steps. Then, in order to address the user/location cold-start issues, we incorporate rich contextual information related to users' check-ins into a recommender system. We propose using different deep neural network-based methods to build recommender systems. The first approach is based on a two-encoder approach, while the second approach is based on the Transformer-based model BERT. The goal is to evaluate the efficacy of various methods for achieving our objectives. To assess the proposed model, we run extensive experiments on real-world data sets using a variety of state-of-the-art baseline methods and evaluation metrics. The experimental results show that our methods are effective for detecting real POIs, with an accuracy of 94 percent, a balanced F1-score of 95 percent, and a Hit Ratio score of around 92 percent during top@50 for the recommendation task.