Crowd-Sourced Label Filtering via Browsing History Delta Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting fraudulent labels in web application services are inadequate in identifying and filtering abnormal labels triggered by external events, leading to resource-intensive server handling of incorrect labeling activities.
Innovation Solution
A method and system that analyze crowd-sourced labels by assessing user browsing history to identify abnormal patterns, separating browsing history into groups, generating a delta set of web resources associated with abnormal visits, and associating these resources as the source of external trigger events, thereby discarding or lowering the weight of labels from users who accessed these resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing methods for detecting fraudulent labels are used, then fraudulent labels can be filtered to some extent, but abnormal labels triggered by external events cannot be identified, leading to resource-intensive server handling
Solution Approach 1:
The system performs preliminary analysis of user browsing history before labels are submitted. By pre-identifying users who have accessed trigger web resources (scandalous reviews, fake news, etc.), the system can proactively flag or discard their labels, preventing resource-intensive processing of abnormal labels later.
Solution Approach 2:
The patent introduces an intermediary detection layer that analyzes user browsing behavior between label submission and label processing. This intermediary system uses web history logs and trigger resource identification to filter abnormal labels, reducing the burden on the main label processing server.
2Quantity of substance
If all crowd-sourced labels are processed without filtering, then label volume is maintained, but incorrect labels from external trigger events reduce labeling accuracy
Solution Approach 1:
The system extracts and removes abnormal labels from the overall label set by identifying users who accessed trigger web resources. This extraction process separates fraudulent/abnormal labels from legitimate ones, maintaining label volume while improving accuracy through selective removal of problematic labels.
Solution Approach 2:
The system implements a feedback mechanism where user browsing history is continuously monitored and analyzed. When trigger resources are detected in user browsing patterns, this feedback information is used to automatically discard or lower the weight of subsequent labels from these users, improving overall labeling accuracy.
3Measurement precision
If user browsing history is analyzed to identify abnormal patterns, then confidence in labeling accuracy increases, but system complexity increases
Solution Approach 1:
The system uses a universal browsing history analysis mechanism that serves multiple functions: detecting fraudulent labels, identifying abnormal external-triggered labels, and tracking user behavior patterns. This multi-functional approach improves labeling accuracy without requiring separate complex systems for each detection task.
Solution Approach 2:
The system leverages existing web history logs and automated pattern recognition algorithms to perform self-service detection of abnormal labels. By using pre-collected browsing data and automated trigger resource identification, the system reduces manual intervention complexity while maintaining high labeling accuracy.
Data Source
AI summary
Systems and methods for determining an abnormal crowd-sourced label for a digital item comprising: analyzing a portion of the plurality of crowd-sourced labels, determining an abnormal subset of crowd-sourced labels having been potentially caused by an occurrence of the external trigger event, acquiring a browsing history associated with a subset of the plurality of users, separating the browsing history into a first browsing history group and a second browsing history group associated, generating a delta set of web resources based on analyzing the first browsing history group and the second browsing history group for differences in web resources visited by the subset of the plurality of users, the delta set containing at least one web resource of the first browsing history set being associated with an abnormal pattern of visits, associating the at least one web resource as being a source of the external trigger event.