Time-Aware Dual LSTM Neural Network with Similarity Graph Learning for Remote Sensing Service Recommendation

Area of Science:

• Earth Science
• Computer Science
• Data Science

Background:

• Advancements in Earth observation and satellite technology have increased remote sensing resources.
• Challenges persist in ensuring the timeliness and relevance of these resources for end-users.
• Effective recommendation systems are needed to navigate vast remote sensing data archives.

Purpose of the Study:

• To develop a precise remote sensing resource service recommendation model.
• To address the limitations of existing systems in providing timely and relevant data.
• To enhance end-user decision-making in specific domains using remote sensing data.

Main Methods:

• Proposed a model combining time-aware dual LSTM neural networks with similarity graph learning.
• Constructed user interaction history sequences and a category similarity graph.
• Utilized Long Short-Term Memory (LSTM) for sequence representation and Graph Convolutional Networks (GCN) for graph structures.
• Incorporated stream push technology and user ID embedding for enhanced modeling.

Main Results:

• The model effectively represents historical sequences and graph structures for accurate recommendations.
• User characteristics and similarity relationships were successfully modeled.
• Experimental evaluation on three datasets demonstrated superior performance compared to state-of-the-art algorithms.

Conclusions:

• The proposed recommendation model significantly improves the precision and relevance of remote sensing service suggestions.
• The integration of time-aware LSTMs and graph learning offers a robust solution for remote sensing data discovery.
• This approach provides a valuable tool for end-users to make informed decisions based on Earth observation resources.