Trust-Enhanced POI Recommendation Algorithm Using Expectation-Maximization

Point-of-interest (POI) recommendation systems have become increasingly important as travelers rely on mobile technologies and location-based social networks to discover new places. However, existing approaches often struggle with static user preferences, inadequate trust modeling, and extreme data sparsity. This paper introduces ExMax, a dynamic trust-enhanced recommendation framework leveraging Expectation-Maximization theory to address these limitations. ExMax employs a novel check-in matrix representation that adapts to evolving user interests, incorporates friendship network information to enhance recommendation quality, and integrates sentiment analysis to capture nuanced satisfaction signals beyond ratings. The
framework’s iterative probabilistic model discovers latent features within sparse data, enabling meaningful recommendations even with limited interaction history. The algorithm exhibits O(T · |R|·F · K) time complexity for offline learning, where T represents EM iterations (typically 20-30), |R| denotes observed ratings, F indicates features, and K represents gradient steps. While sparse matrix operations and parallelization potential provide some mitigation, the iterative nature poses scalability challenges for platforms with hundreds of millions of users. Experimental evaluation on Yelp, Gowalla, and Brightkite datasets demonstrates that ExMax performs favorably compared to existing approaches across various metrics. The results suggest that integrating dynamic preference modeling, social trust signals, and contextual information offers a promising direction for location-based recommendation systems, particularly where recommendation quality justifies the computational cost. This work demonstrates how probabilistic modeling can effectively capture the dynamic and social nature of location discovery while acknowledging the inherent computational trade-offs of iterative optimization.