Explainable Multihop Social Link Prediction Based on Temporal Logical Rules in Dynamic Social Networks.

Area of Science:

• Social Network Analysis
• Artificial Intelligence
• Data Mining

Background:

• Social link prediction is crucial for understanding user interactions in dynamic networks.
• Existing embedding methods struggle with explainability, multirelations, and multihop predictions.
• Dynamic social networks require models that capture temporal evolution.

Purpose of the Study:

• To propose an innovative multihop temporal social link prediction model.
• To address limitations in explainability, multirelations, and multihop relation prediction.
• To leverage temporal knowledge graphs and logic rules for improved link prediction.

Main Methods:

• Constructing Temporal Social Knowledge Graphs (TSKGs) from dynamic social networks.
• Incorporating orthogonal transformation matrices into Graph Neural Networks (GNNs) for time-aware representations.
• Defining temporal social random walks to generate temporal social rules and using Temporal Logic Embedding (TLE) for prediction.

Main Results:

• TLE effectively models dynamic social networks using TSKGs.
• The model demonstrates superior performance in social link prediction across four datasets.
• TLE provides explainability for multihop link predictions by combining confidence scores and time differences.

Conclusions:

• TLE offers a novel and effective approach for multihop temporal social link prediction.
• The model overcomes key limitations of previous embedding-based methods.
• TLE enhances the understanding and prediction of evolving social network structures.