Device Similarity Scoring via Statistical Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for device similarity analysis, such as rule-based systems, are inflexible and require human tuning, struggling to control false positive and false negative rates, and are ineffective in adapting to changing device element frequencies and missing data, especially in web-based applications like e-commerce where secure device identification is crucial.
Innovation Solution
A data-driven approach using a statistical modeling framework, specifically a Naïve Bayesian model with Expectation-Maximization algorithm, to calculate a soft similarity score between devices based on their components, automatically learning from unlabeled data and adapting to element importance and dynamics, without relying on human intervention or explicit labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based systems are used for device similarity analysis, then the system structure is simple and easy to implement, but the system is inflexible and cannot adapt to changing device element frequencies
Solution Approach 1:
The patent replaces rule-based mechanical systems with a statistical modeling framework using Naïve Bayesian models and Expectation-Maximization algorithms. This substitution enables automatic learning from data to determine device similarity, eliminating the need for manual rule tuning while adapting to changing device element frequencies through probabilistic calculations.
2Reliability
If manual tuning of rule-based systems is performed, then initial system setup is possible, but continuous human intervention is required to control false positive and false negative rates
Solution Approach 1:
The patent implements self-service through automatic learning mechanisms where the Naïve Bayesian model and Expectation-Maximization algorithm continuously optimize device similarity determination without human intervention. The system automatically adjusts to control false positive and false negative rates by learning from observed device element frequencies and patterns in the data.
Solution Approach 2:
The patent incorporates feedback loops where the system continuously monitors device element frequencies and similarity determination outcomes, using this feedback to refine probabilistic models and improve accuracy over time. The Expectation-Maximization algorithm utilizes feedback from unlabeled data to iteratively improve model parameters and control error rates.
3Ease of operation
If conventional device identification methods are used, then implementation is straightforward, but the system cannot handle missing data effectively
Solution Approach 1:
The patent changes the approach from deterministic rule-based parameter matching to probabilistic parameter assessment. The Naïve Bayesian model handles missing data by calculating probabilities based on available device elements, allowing the system to determine device similarity reliably even when some device components are missing or unavailable.
4Device complexity
If cookies or explicit tagging are used for device identification, then device tracking is simple, but security vulnerabilities increase due to cookie theft and replay attacks
Solution Approach 1:
The patent extracts the device identification function from cookie-based systems and implements it through analysis of intrinsic device elements. By removing the cookie dependency and using statistical modeling of device characteristics, the system eliminates security vulnerabilities associated with cookie theft and replay attacks while maintaining device identification capability.
Data Source
AI summary
A method is used in analyzing device similarity. Data describing a device is received and a model is applied to the data. Based on the modeling, a measure of similarity between the device and a previously known device is determined.