Variable Sliding-Window Routine Mining for Analyst-Focused Automation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing routine discovery systems face challenges in efficiently identifying important automation routines without overwhelming analysts with excessive findings and maintaining reasonable report runtime, particularly in front office and back office operations.
Innovation Solution
A sliding window routine mining system that mines automation routines directly from a long stream of operator actions without segmentation, using varying window sizes and addressing conflicts through filtering and conflict resolution methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional routine discovery systems segment the action stream into sentences to identify individual repetitive tasks, then they can isolate specific tasks, but they generate excessive findings that overwhelm analysts and increase computational overhead
Solution Approach 1:
The patent extracts only the most significant automation routines from the action stream by applying frequency thresholds and significance scoring, rather than presenting all possible segmented tasks. This filters out noise while preserving important automation opportunities, directly addressing the problem of overwhelming analysts with excessive findings.
Solution Approach 2:
The system changes parameters by varying window sizes (e.g., 5, 10, 15 actions) and adjusting frequency thresholds to optimize the balance between detection precision and computational efficiency. This allows the system to adapt to different task complexities while maintaining reasonable analysis performance.
2Adaptability or versatility
If the system scans with multiple window sizes to find patterns, then it improves routine discovery comprehensiveness, but it increases report runtime and computational complexity
Solution Approach 1:
The system performs preliminary filtering by action frequency and significance scoring before applying the full multi-window scanning process. This preliminary action reduces the search space and allows the system to focus computational resources on promising patterns, thereby maintaining comprehensive detection while reducing overall runtime.
Solution Approach 2:
The system dynamically adjusts the window size range and scanning depth based on the characteristics of the action stream and preliminary analysis results. This dynamic adaptation allows comprehensive pattern detection when needed while reducing computational overhead for simpler cases, directly addressing the runtime vs. comprehensiveness trade-off.
3Reliability
If the system presents all discovered routines to analysts, then it ensures no important routines are missed, but it overwhelms analysts with too many findings
Solution Approach 1:
The system applies different quality criteria and filtering thresholds to different types of routines based on their characteristics. High-frequency, high-significance routines receive preferential treatment in the presentation hierarchy, while less significant routines are filtered or grouped. This local quality differentiation ensures important routines are highlighted without presenting every possible finding to the analyst.
Solution Approach 2:
The system introduces an intermediary significance scoring mechanism that acts as a mediator between the raw pattern detection and the analyst presentation. This intermediary layer ranks and filters routines based on multiple criteria (frequency, window size, action diversity), providing a curated subset that maintains reliability while reducing analyst workload.
Data Source
AI summary
A system is adapted to automatically create new automation routines. The system includes a processor configured to, over a period of time, store L actions taken by one or more customer support agents and, in real time, for values of n between a first loop value and a second loop value: with a window of length n, starting at the first position within the stored actions and ending at the L-nth position within the stored actions, repeatedly: scan the stored L actions; create an automation candidate from the actions that fall within the window; store the automation candidate; increment the position of the window; and increment the value of n. The processor is further configured to identify repeated automation candidates of the stored automation candidates; filter the repeated automation candidates to select final automation candidates; and create new automation routines from the final automation candidates.