Dark Web Forum Labeling via Semi-Supervised Tag Hierarchy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies for classifying dark web forum discussions face challenges due to labeled data scarcity and imbalanced data classes, which are time-consuming and inefficient, and fail to address the need for a more granular and versatile classification method.
Innovation Solution
A computer-implemented system that automatically assigns hierarchical labels to dark web forum topics using a four-module approach: crawling and data extraction, preprocessing, feature extraction with Doc2vec, and multi-class, multi-label classification with machine classifiers, incorporating techniques like SMOTE and Elastic Search for addressing class imbalance and leveraging tag hierarchies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling of training data is used, then label accuracy can be ensured, but the process becomes time-consuming and un-scalable
Solution Approach 1:
The patent uses semi-supervised learning where a small set of manually labeled data is copied and propagated to label large amounts of unlabeled data through confidence-based propagation. High-confidence predictions from the model are copied as labels to expand the training set without manual intervention, resolving the contradiction between accuracy and time consumption.
Solution Approach 2:
The system performs self-labeling by using its own predictions to create training data. The model predicts labels for unlabeled data, and high-confidence predictions are automatically added to the training set, allowing the system to serve itself and eliminate the need for continuous manual labeling.
2Device complexity
If traditional classification models are used, then implementation simplicity is maintained, but they fail to address labeled data scarcity and imbalanced data classes
Solution Approach 1:
The patent segments the classification problem into multiple stages: initial supervised learning on labeled data, semi-supervised learning on unlabeled data, and hierarchical label assignment. This segmentation allows each stage to address specific challenges (data scarcity, class imbalance) while maintaining overall system manageability.
Solution Approach 2:
The patent combines multiple learning approaches (supervised learning, semi-supervised learning, SMOTE for oversampling) into a composite classification system. This composite approach leverages the strengths of each method to address both data scarcity and class imbalance while maintaining reasonable complexity.
3Adaptability or versatility
If broad categorization is used, then data coverage is comprehensive, but granular classification of discussion topics is insufficient
Solution Approach 1:
The patent introduces a hierarchical dimension to the classification system, organizing labels into parent-child relationships. This adds a structural dimension to the classification, allowing broad categories to be maintained at higher levels while enabling granular classification at lower levels, thus addressing both comprehensive coverage and fine-grained categorization needs.
Solution Approach 2:
The patent segments the label space into hierarchical levels (parent labels and child labels), allowing the system to operate at different granularities as needed. This segmentation enables versatile classification by permitting users to query at either broad or specific levels without being constrained by a single classification depth.
Data Source
AI summary
Embodiments of a computer-implemented system for improving classification of data associated with the deep web or dark net are disclosed.