User tendency-based rating scaling in online trading networks

Area of Science:

• Social Network Analysis
• Machine Learning
• Data Mining

Background:

• User ratings in social networks vary due to subjective scoring criteria.
• Accurate user classification is crucial for trust and security in online platforms, especially anonymous ones.
• Existing rating systems struggle to differentiate user satisfaction levels consistently.

Purpose of the Study:

• To propose and evaluate a novel user tendency-based rating scaling method.
• To enhance the classification accuracy of "good" and "bad" users in online trade social networks.
• To improve performance in user classification, link prediction, and clustering tasks.

Main Methods:

• Developed a user tendency-based rating scaling algorithm.
• Applied weighted graph embedding using original and scaled ratings.
• Evaluated performance on three real-world online rating network datasets.

Main Results:

• The proposed rating scaling method significantly outperformed original ratings.
• Achieved up to 17% improvement in classification accuracy.
• Showed up to 2.5% improvement in link prediction (AUC ROC) and 21% in clustering (Dunn-index).

Conclusions:

• User tendency-based rating scaling effectively improves user classification and network analysis.
• The method enhances the reliability of reputation systems in online trade.
• This approach offers a more nuanced understanding of user behavior and network dynamics.