Maximizing Influence in Social Networks Using Combined Local Features and Deep Learning-Based Node Embedding.

Area of Science:

• Network Science
• Computer Science
• Data Mining

Background:

• Influence maximization is crucial for viral marketing and information diffusion but faces challenges like low infection rates and high time complexity in large networks.
• Existing methods often struggle with scalability due to computational demands or reliance on free parameters, limiting their practical application.
• Addressing these limitations is essential for developing efficient influence maximization strategies in complex, real-world networks.

Purpose of the Study:

• To propose a novel local heuristic algorithm, the Embedding Technique for Influence Maximization (ETIM), designed to overcome the limitations of existing influence maximization methods.
• To enhance the efficiency and effectiveness of influence maximization in large-scale networks by reducing search space and computational overhead.
• To improve infection rates and solution quality compared to existing algorithms.

Main Methods:

• ETIM employs shell decomposition, graph embedding, and reduction, integrating local structural features for candidate node selection.
• A deep learning-based node embedding technique generates multidimensional vectors for candidate nodes, capturing complex network relationships.
• Node dependency on spreading is calculated using local topological features, followed by identification of influential nodes based on combined local and embedded features.

Main Results:

• ETIM demonstrates competitiveness and achieves superior performance in solution quality when evaluated using the independent cascade model.
• The algorithm significantly reduces computational overhead and search space by focusing on network shells and topological features.
• ETIM achieves a substantial improvement in infection rate, averaging 12% higher than the collective influence global algorithm, while being considerably faster.

Conclusions:

• The Embedding Technique for Influence Maximization (ETIM) offers an efficient and effective solution for the influence maximization problem in large-scale networks.
• ETIM's hybrid approach, combining graph embedding with local structural analysis, successfully addresses the challenges of time complexity and solution quality.
• The proposed method presents a promising advancement for applications requiring efficient identification of influential nodes in complex network structures.