User Estimation via Page Transition and Time Interval Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing user estimation techniques struggle to accurately identify users based on browsing behavior patterns when the number of pages browsed or links included is small, leading to low estimation accuracy.
Innovation Solution
A user estimation apparatus that extracts features such as the order of page transitions and time intervals between page visits, creating a model for each user and using it to estimate user identity, even with limited browsing data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional user estimation techniques using browsing behavior patterns are applied, then user identification can be performed, but estimation accuracy deteriorates when the number of pages browsed or links included is small
Solution Approach 1:
The patent transforms the browsing behavior data into a standardized feature vector with fixed dimensionality. By converting variable-length browsing sequences into fixed-dimensional features (including page transition counts, time interval statistics, and browsing depth metrics), the system enables accurate user estimation even when the original number of pages browsed is small. This parameter transformation allows the machine learning model to process inconsistent data lengths uniformly.
Solution Approach 2:
The patent enriches the feature space by extracting multiple dimensions of browsing behavior characteristics from limited page visits. Instead of relying solely on the raw number of pages browsed, the system extracts features across multiple dimensions including transition frequency, time interval distributions, browsing depth, and behavioral patterns, thereby creating a more informative feature representation that improves estimation accuracy.
2Measurement precision
If conventional user estimation techniques using browsing behavior patterns are applied, then user identification can be performed, but estimation accuracy deteriorates when the number of links included in the page is small
Solution Approach 1:
The patent converts the limited link information into standardized feature vectors that capture the essential characteristics of browsing behavior. By transforming the raw link data into fixed-dimensional features including link transition counts, time interval statistics, and behavioral patterns, the system achieves accurate user estimation even when the original number of links is small.
Solution Approach 2:
The patent enriches the feature representation by extracting multiple dimensions of behavioral characteristics from limited link data. The system analyzes browsing patterns across multiple feature dimensions including transition frequency, temporal patterns, and navigation behavior, thereby creating a comprehensive feature set that improves estimation accuracy beyond what raw link counts provide.
Data Source
AI summary
A user estimator includes an extractor extracting at least either order of page transitions on a website by a user or a time interval of transition to each page, as a feature amount of page browsing by the user, from data to be learned and representing a request by the user to the website, and extracting at least either order of page transitions on the website or a time interval of transition to each page, as a feature amount of page browsing by any user, from data to be estimated and representing requests by the users to the website, a learning unit creating a model indicating a feature of page browsing for each user, by learning the extracted feature amount, to be learned, of page browsing by each user, and an estimation unit referring to the feature amount, to be estimated, and the model, and estimating the user among users.