Graph Structure Model Training for Junk Account Identification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for identifying junk accounts on online platforms are unreliable, as they lack effective solutions for distinguishing between normal and abnormal account behaviors, particularly for newly registered accounts that do not have sufficient account portrait information.

Innovation Solution

A graph structure model training method that utilizes an account medium network graph, featuring nodes representing accounts and media with edges indicating login behavior relationships, along with feature data and risk labeling data, to identify junk accounts through supervised graph embedding, leveraging media aggregation and time aggregation characteristics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional junk account identification methods are used, then the identification process is simple, but the reliability of junk account detection is low

Engineering Contradiction:
Improvereliability of junk account detectionVSAvoidcomplexity of identification system
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent transforms the junk account identification problem from traditional feature-based analysis to graph structure embedding space. By mapping accounts, devices, and IPs into a unified graph embedding space, the system captures complex relational patterns across multiple dimensions (account-device relationships, account-IP relationships, device-IP relationships) that traditional methods cannot detect, thereby improving detection reliability without linearly increasing system complexity

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent introduces graph embedding technology as an intermediary layer between raw data and junk account identification. The embedding model acts as a mediator that transforms complex relational data into compressed feature representations, enabling the identification system to capture subtle patterns while maintaining computational efficiency. This intermediary transformation resolves the contradiction by providing high-reliability detection without requiring proportionally complex processing

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If more account portrait information is collected to improve identification accuracy, then the reliability improves, but the loss of time for data collection increases

Engineering Contradiction:
Improveidentification accuracyVSAvoidtime for data collection
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-training graph embedding models on historical account behavior data, device information, and IP relationship data. This pre-training establishes baseline embeddings that can be quickly updated with new data. When identifying new accounts, the system leverages pre-computed embeddings for devices and IPs, avoiding the need to collect and process all possible account portrait information from scratch, thus reducing time loss while maintaining high identification accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes the parameter representation from traditional account portrait features to graph embedding vectors. These embedding parameters capture essential relational information in a compressed form, allowing the system to achieve high identification accuracy with fewer data collection requirements. The embedding dimension and aggregation depth can be adjusted to balance between accuracy and data collection time

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If graph structure model with comprehensive relationships is used, then the measurement precision of abnormal behavior patterns improves, but the device complexity increases

Engineering Contradiction:
Improveprecision of abnormal behavior detectionVSAvoidcomplexity of graph structure model
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the graph structure model into distinct components: account embedding module, device embedding module, IP embedding module, and relationship aggregation module. Each component handles specific types of data and relationships independently. This segmentation allows the system to achieve high measurement precision by comprehensively analyzing multiple relationship types while managing complexity through modular architecture, where each segment can be optimized and maintained separately

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3703332B1Graph structure model training and JUNK account identification
Publication Date: 2021.11.10 ADVANCED NEW TECHNOLOGIES CO LTD
  • EP3703332B1 patent drawingFigure 1~2
  • EP3703332B1 patent drawingFigure 3~4
  • EP3703332B1 patent drawingFigure 5~6

AI summary

Implementations of the present specification disclose graph structure model training and junk account identification methods, apparatuses, and devices. The solution includes: obtaining an account medium network graph, a node in the account medium network graph representing an account and a medium, and at least some edges indicating that a login behavior relationship exists between nodes connected by the edges; obtaining feature data and risk labeling data of the node, the feature data reflecting a login behavior of the corresponding node in a time sequence; and training, based on the account medium network graph, the feature data, and the risk labeling data, a predefined graph structure model for identifying a junk account.