Integrating user reviews and risk factors from social networks in a multi-objective recommender system

With the rapid expansion of location-based social networks, the significance of recommendations for tours and activities has grown significantly. However, many recommender systems overlook the valuable insights embedded in user reviews, thereby missing out on potential improvements. By incorporating review text into the recommendation process, it is possible to enhance the performance of recommender systems and tackle the Cold Start problem. In this study, a novel personalized method for recommending Points of Interest (POI) trips based on user reviews is proposed. The approach leverages a transformer model to extract semantic correlations in a multitask-learning setting, enabling predictions on crime rates and traffic flow at various POIs. Additionally, an LSTM network is employed to identify similar users based on their feedback, thereby overcoming data scarcity and the Cold Start issue. Moreover, this method incorporates multifaceted contextual information to determine user preferences accurately. Finally, a neural hybrid framework mines personalized POIs in a sequential pattern and integrates them into the recommendation process to identify the most efficient trip candidates. The proposed methodology is evaluated using datasets from Yelp, Gowalla, and Tripadvisor, and its superior performance is demonstrated across multiple evaluation metrics, including MAP, NDCG, RMSE, and F-Score.