The evolution of social networks has given rise to significant challenges associated with the overwhelming amount of information available. These challenges encompass various areas such as viral marketing, disease management, and misinformation control. Crafting effective strategies for minimizing influence is heavily influenced by factors like network topology, user behavior, and the dynamics of information propagation. As social networks become more intricate, the imperative to utilize data-driven insights becomes increasingly apparent. The Social Influence Minimization Problems (IMIN) aims to identify and strategically block users to limit the spread of information. Extracting structural insights through data-mining techniques can guide the development of efficient heuristics and the identification of influential users to be targeted for blocking. To address the NP-hard nature of the IMIN problem, a robust metaheuristic algorithm based on the Greedy Randomized Adaptive Search (GRASP) framework has been introduced. This method is derived from a deep understanding of how network features contribute to impactful solutions, proving to be effective and cost-efficient when compared to state-of-the-art methods.

• Journal: European Journal of Operational Research
• Impact Factor: 6.0 (2024)
• Paper link: TODO
• Area: Operations Research & Management Science
• Quartil: Q1 - 14/106 (2024)

The experiments use real-world social network datasets with two propagation model variants (1 and 2) for each network.

Each instance file follows this structure:

1. First line: propagation_model num_infected_nodes num_blockers seed

   • Propagation model identifier:
     • 1 = WCM (Weighted Cascade Model): Probabilistic propagation based on edge weights
     • 2 = TV (Trivalency): Propagation model with three-state node activation
   • Number of seed nodes to be influenced (10)
   • Number of blockers to select (20)
   • Seed for random number generation

2. Second line: List of initially infected node IDs (space-separated)

   • These nodes are calculated according to state-of-the-art SNIMP methods for each propagation model (WCM or TV)

3. Third line: num_nodes num_edges

   • Total number of nodes in the network
   • Total number of edges in the network

4. Remaining lines: Edge list with influence probabilities

   • Format: source_node target_node weight
   • Each line represents a directed edge with its influence probability (used for TV model)

All txt format instances can be found in the instances folder.

• src/: Source code of the GRASP-based metaheuristic proposal for IMIN
  • src/main/java/es/urjc/etsii/grafo/IMIN/: Main package containing models, algorithms, constructives, and experiments
• instances/: Contains 16 social network instance files in SNIMP format (8 networks × 2 propagation models)
• results.xlsx: Experimental results including objective values, computational times, and comparisons with state-of-the-art methods
• pom.xml: Maven project configuration file
• IMIN-0.17.jar: Executable JAR file of the application (see Executable section below)

You can just run the IMIN-0.17.jar as follows (the algorithm will select manuscript parameters).

java -jar IMIN-0.17.jar

Once you run the algorithm, the screen will show you the metrics and status. You can also found a folder named as experiments where an excel will generated when the experiment end.

Please cite our paper if you use it in your own work:

Bibtext

MDPI and ACS Style

AMA Style

Chicago/Turabian Style