Two-Stage Classifier for Human Interaction Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for distinguishing genuine human interaction from non-genuine human interaction, such as bot activity, are inconvenient for users and lack effectiveness in identifying random or software-controlled interactions.
Innovation Solution
A computer-implemented method using two classifiers to determine whether interaction data indicates human or random interaction, and human or software-controlled interaction, based on parameters like movement efficiency, timing entropy, and UI cues, to accurately identify human interaction and detect replayed interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If CAPTCHA or challenge tests are provided to distinguish human from bot interaction, then the ability to identify non-genuine interaction is improved, but the convenience for genuine human users deteriorates
Solution Approach 1:
The system performs automatic classification of interaction data using machine learning models without requiring users to complete manual challenges. The classification is done self-service style by analyzing interaction patterns, timing, and behavior characteristics, eliminating the need for CAPTCHA completion while maintaining security
Solution Approach 2:
The patent replaces the mechanical CAPTCHA challenge system with an automated machine learning-based classification system. Instead of requiring users to solve puzzles or complete challenges, the system automatically analyzes interaction data patterns to distinguish human from bot interactions in the background
2Reliability
If behavioral biometrics like mouse and keystroke dynamics are used to distinguish human from bot, then the detection capability is improved, but the system complexity increases
Solution Approach 1:
The system segments the classification task into two distinct classifiers: a first classifier that determines whether interaction is directed or random, and a second classifier that determines whether interaction is human or software-controlled. This segmentation allows each classifier to specialize in specific aspects of interaction analysis, improving detection capability while managing complexity through modular design
Solution Approach 2:
The patent adds a new dimension to interaction analysis by introducing a two-stage classification approach with different analysis dimensions. The first classifier analyzes interaction directionality (directed vs random), while the second classifier analyzes interaction origin (human vs software). This multi-dimensional approach enhances detection capability without requiring a single overly complex system
3Measurement precision
If a two-classifier system is used to determine human or random interaction and human or software-controlled interaction, then the accuracy in identifying human interaction is improved, but the computational processing time increases
Solution Approach 1:
The first classifier performs a preliminary assessment of interaction data to determine whether the interaction is directed or random. This preliminary action filters out obviously random interactions early in the process, allowing the second classifier to focus only on directed interactions that require more sophisticated analysis, thereby reducing overall processing time while maintaining high accuracy
Data Source
Figure 1~2
Figure 3~5
Figure 4
AI summary
Embodiments of the present invention provide a computer-implemented method of identifying human interaction with a computer, comprising receiving interaction data indicative of interaction with the computer, determining, at a first classifier, whether the interaction data is indicative of directed or random interaction with the computer, determining, at a second classifier, whether the interaction data is indicative of human or software-controlled interaction with the computer, and determining whether the interaction data is indicative of human interaction with the computer based on an output of at least one of the first and second classifiers.