Semisupervised Network Embedding With Differentiable Deep Quantization.

Area of Science:

• Computer Science
• Network Science
• Machine Learning

Background:

• Learning low-dimensional network embeddings is vital for network analytics.
• Large networks pose storage and processing challenges due to embedding size.

Purpose of the Study:

• To develop a method for compressing network embeddings.
• To reduce storage footprint and accelerate retrieval speed for large network embeddings.

Main Methods:

• Introduced d-SNEQ, a differentiable DNN-based quantization method for network embedding.
• Incorporated a rank loss to preserve high-order information in quantization codes.
• Proposed a new evaluation metric, path prediction, for assessing high-order information preservation.

Main Results:

• d-SNEQ substantially compresses trained embeddings, reducing storage and improving retrieval speed.
• Achieved superior performance in link prediction, path prediction, node classification, and node recommendation.
• Demonstrated significant space- and time-efficiency compared to state-of-the-art methods.

Conclusions:

• d-SNEQ offers an effective solution for efficient network embedding storage and retrieval.
• The method preserves crucial high-order network information, enhancing downstream task performance.
• d-SNEQ presents a practical advancement for large-scale network analysis.