Intelligent Sensors for POI Recommendation Model Using Deep Learning in Location-Based Social Network Big Data

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Existing Point of Interest (POI) recommendation models struggle to extract deep feature information from social network big data.
• Location-Based Social Networking (LBSN) generates vast amounts of data, necessitating efficient feature extraction for accurate recommendations.

Purpose of the Study:

• To propose a novel POI recommendation model leveraging deep learning for enhanced feature extraction in social network big data.
• To improve the accuracy and recall of POI recommendation systems by effectively utilizing geographic, semantic, and temporal features.

Main Methods:

• A POI static feature extraction method using symmetric matrix decomposition to capture geographical and category features.
• An improved Continuous Bags-of-Words (CBOW) model for extracting semantic features from user comments, creating implicit vector representations.
• A Geographical-Spatiotemporal Gated Recurrent Unit Network (GSGRUN) to learn user preferences from check-in history and distinguish individual check-in behaviors.

Main Results:

• The proposed GSGRUN model achieved a precision of 0.0686 and recall of 0.0769 on the loc-Gowalla dataset.
• On the loc-Brightkite dataset, the model yielded a precision of 0.0659 and recall of 0.0835.
• Both datasets showed superior performance compared to existing recommendation methods, particularly with a recommendation list length of 15.

Conclusions:

• The developed deep learning-based POI recommendation model significantly improves recommendation system performance.
• The GSGRUN model effectively extracts deep features and learns user preferences, outperforming comparative methods.
• This approach offers a robust solution for accurate POI recommendations in big data environments.