<p>Personalized recommendation in location-based services (LBS) is a challenging task due to the diversity of user preferences and the dynamic nature of location-based interactions. Existing methods often struggle to capture the complex interplay of user interests, leading to suboptimal recommendations. Addressing this problem is crucial for enhancing user experience and driving engagement in LBSs. Accurate and personalized point of interest (POI) recommendations can help users discover relevant places, optimize their time, and make informed decisions, because these points can play the role of the road landmarks. By improving POI recommendations, we can foster a more enjoyable and valuable experience for LBSN users. This paper presents a novel approach to enhancing LBS by leveraging deep learning (DL) techniques and data from location-based social networks (LBSNs) like Foursquare and Yelp. Our method effectively extracts user interests from LBSN data, incorporating individual preferences, social influence, and temporal dynamics. By combining these factors, we provide personalized POI recommendations that significantly improve user experience. We introduce an improved long short-term memory (LSTM) model to capture complex user behavior patterns and extract relevant POIs. Additionally, we address the challenge of data inconsistency across different social networks by harmonizing place categories. In the proposed model, the improved artificial bee colony (IABC) optimally updates the weights, biases, learning rate, and the number of units of the LSTM. This paper also introduces a new three-dimensional migration model to enhance the exploitation capabilities of the ABC algorithm. The performance of the proposed IABC-LSTM model is compared to several advanced machine learning (ML) algorithms, including recurrent neural networks (RNN), K-nearest neighbors (KNN), support vector machines (SVM), biogeography-based optimization (BBO), and orchard algorithm (OA). To evaluate our model, we utilized a rigorous evaluation framework, incorporating metrics such as root mean square error (RMSE), coefficient of determination (R²), convergence curve, precision, recall, F-score, and receiver operating characteristic (ROC) curve. Our research demonstrates the effectiveness of our approach by achieving the highest <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:{R}^{2}\)</EquationSource> </InlineEquation> value of 96.41% on the validation set, and the lowest RMSE value of 0.84, whereas SVM model achieved the highest RMSE and the lowest<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\:\:{R}^{2}\)</EquationSource> </InlineEquation>. Also, the combination of all three factors achieved the highest F-score of 0.893. Our findings highlight the importance of considering multiple factors for personalized recommendations and the potential of DL techniques to revolutionize LBS.</p>

Capturing user dynamics: A novel evolutionary deep learning framework for personalized POI recommendations in navigation

期刊论文
发表日期：

2026年3月10日

查看全文

Hossein Aminzadeh,
Mehrdad Kaveh,
Diego Martín

摘要

Personalized recommendation in location-based services (LBS) is a challenging task due to the diversity of user preferences and the dynamic nature of location-based interactions. Existing methods often struggle to capture the complex interplay of user interests, leading to suboptimal recommendations. Addressing this problem is crucial for enhancing user experience and driving engagement in LBSs. Accurate and personalized point of interest (POI) recommendations can help users discover relevant places, optimize their time, and make informed decisions, because these points can play the role of the road landmarks. By improving POI recommendations, we can foster a more enjoyable and valuable experience for LBSN users. This paper presents a novel approach to enhancing LBS by leveraging deep learning (DL) techniques and data from location-based social networks (LBSNs) like Foursquare and Yelp. Our method effectively extracts user interests from LBSN data, incorporating individual preferences, social influence, and temporal dynamics. By combining these factors, we provide personalized POI recommendations that significantly improve user experience. We introduce an improved long short-term memory (LSTM) model to capture complex user behavior patterns and extract relevant POIs. Additionally, we address the challenge of data inconsistency across different social networks by harmonizing place categories. In the proposed model, the improved artificial bee colony (IABC) optimally updates the weights, biases, learning rate, and the number of units of the LSTM. This paper also introduces a new three-dimensional migration model to enhance the exploitation capabilities of the ABC algorithm. The performance of the proposed IABC-LSTM model is compared to several advanced machine learning (ML) algorithms, including recurrent neural networks (RNN), K-nearest neighbors (KNN), support vector machines (SVM), biogeography-based optimization (BBO), and orchard algorithm (OA). To evaluate our model, we utilized a rigorous evaluation framework, incorporating metrics such as root mean square error (RMSE), coefficient of determination (R²), convergence curve, precision, recall, F-score, and receiver operating characteristic (ROC) curve. Our research demonstrates the effectiveness of our approach by achieving the highest \(\:{R}^{2}\) value of 96.41% on the validation set, and the lowest RMSE value of 0.84, whereas SVM model achieved the highest RMSE and the lowest \(\:\:{R}^{2}\) . Also, the combination of all three factors achieved the highest F-score of 0.893. Our findings highlight the importance of considering multiple factors for personalized recommendations and the potential of DL techniques to revolutionize LBS.