Sybil Detection via Random Sampling Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing collective classification methods fail to accurately detect Sybil accounts in social networks and online platforms, especially when adversarial attacks manipulate graph structures to evade detection.
Innovation Solution
A random sampling-based collective classification method that involves performing a first collective classification, constructing new training data by randomly sampling nodes, and applying posterior score differences to adjust prior scores iteratively, thereby enhancing detection accuracy of Sybil nodes even under adversarial conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing collective classification method is used, then classification process is simple, but detection accuracy of Sybil accounts under adversarial attack is low
Solution Approach 1:
The classification process is divided into multiple iterations, where in each iteration a subset of nodes is randomly sampled for reclassification. This segmentation allows the system to focus computational effort on specific portions of the graph while maintaining overall detection accuracy, resolving the contradiction between accuracy and complexity.
Solution Approach 2:
Instead of reclassifying all nodes in each iteration, the method applies partial action by randomly sampling only a portion of nodes for reclassification. This reduces computational complexity while still improving detection accuracy through multiple passes, effectively balancing the trade-off between accuracy and process complexity.
2Measurement precision
If random sampling-based collective classification is performed, then detection accuracy of Sybil nodes is improved, but computational complexity increases
Solution Approach 1:
The method performs partial reclassification by randomly sampling a subset of nodes in each iteration rather than processing the entire graph. This partial action maintains improved detection accuracy while significantly reducing the computational power required compared to full graph reclassification in each iteration.
Solution Approach 2:
The classification process is executed periodically in multiple iterations, where each iteration applies random sampling and reclassification. This periodic execution allows the system to progressively improve detection accuracy over time while distributing computational load, thereby managing computational power requirements effectively.
3Reliability
If multiple collective classifications are performed with random sampling, then robustness against adversarial attacks is enhanced, but processing time increases
Solution Approach 1:
By performing partial reclassification on randomly sampled node subsets rather than the entire graph, the method enhances robustness against adversarial attacks through multiple iterations while minimizing the time lost to processing. The partial action approach ensures that adversarial manipulations are detected across different samples without requiring exhaustive processing of all nodes.
Solution Approach 2:
The method performs preliminary random sampling and reclassification in multiple iterations before final classification. This preliminary action on sampled data helps identify and correct adversarial manipulations early in the process, improving robustness while reducing the time required for final classification compared to exhaustive methods.
Data Source
AI summary
Disclosed is a random sampling-based collective classification method and system for Sybil account detection. A random sampling-based collective classification method performed by a collective classification system may include performing a first collective classification using training data; constructing sampled new training data by randomly extracting a portion of the entire nodes based on a label assigned to each node according to a result of performing the first collective classification; performing a second collective classification using the constructed new training data; and applying a posterior score difference of each node computed through the first collective classification and the second collective classification to a prior score of each node to be used for training data at a next iteration.