Task Discovery From Computer Usage Data With Privacy-Aware Event Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional user monitoring applications fail to accurately identify specific tasks performed by users on computers, capture sensitive personal information, and provide noisy data due to interleaved tasks, leading to ineffective task analysis and potential security risks.
Innovation Solution
A system that collects computer usage data by capturing user actions and contextual information, clusters events to identify repeated sequences, and generates metrics for automating tasks, while avoiding sensitive information capture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional monitoring applications capture all user actions continuously, then complete task data is obtained, but data quality deteriorates due to noisy interleaved tasks and personal information capture
Solution Approach 1:
The patent extracts only the essential task-related information from the continuous stream of user actions. Instead of capturing all keystrokes and mouse movements, the system identifies and extracts specific task-defining actions and their contextual information, filtering out noisy interleaved tasks and personal information in the process.
Solution Approach 2:
The system performs preliminary analysis of user actions to identify task boundaries and characteristics before final task identification. By pre-processing the action stream to recognize task-starting actions and contextual patterns, the system improves subsequent task identification accuracy while reducing noise.
2Measurement precision
If monitoring applications capture detailed contextual information, then task identification accuracy improves, but security risks increase due to personal information capture
Solution Approach 1:
The patent converts the potential harm of capturing personal information into a benefit by using contextual information strategically. The system captures contextual information (application names, window titles, action sequences) that is sufficient for task identification while deliberately excluding sensitive personal data, thus turning a security risk into an effective task analysis tool.
Solution Approach 2:
The system applies different quality levels of information capture to different aspects of task monitoring. High-quality detailed capture is applied to task-relevant contextual information (application context, action sequences), while personal information is either not captured or captured at minimal quality levels, creating a differentiated information collection strategy.
3Loss of information
If monitoring applications analyze all captured data, then comprehensive task analysis is achieved, but processing time and computational resources increase
Solution Approach 1:
The patent extracts only the essential features and patterns needed for task identification from the captured action data. By focusing on key task-defining actions and their immediate contextual information rather than analyzing every single user interaction, the system achieves comprehensive task analysis with reduced processing requirements.
Solution Approach 2:
The system performs partial analysis on the full data stream by identifying and analyzing only the portions of user actions that are relevant to task identification. Instead of uniformly processing all captured data, the system selectively analyzes task-critical segments, achieving sufficient task analysis completeness with reduced computational effort.
Data Source
AI summary
According to at least one aspect, a system for analyzing computer usage data of a user to identify an underlying task being performed by the user is provided. The system includes a hardware processor and a non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by the hardware processor, cause the hardware processor to perform: receiving a plurality of events each indicative of an action performed by a user and contextual information associated with the action performed by the user, clustering the plurality of events to generate a plurality of clustered events, identifying a plurality of sub-tasks in the plurality of clustered events that each comprise a sequence of clustered events, identifying a task in the plurality of clustered events being performed by the user that comprises at least one sub-task, and generating a score for the task indicative of a difficulty of automating the task.