Abstract
Field Computer-implemented systems for detecting employee knowledge deficits by analyzing the entropy of user interface interaction patterns during enterprise software workflows. Background Employees who lack knowledge in a particular area tend to exhibit uncertainty when performing related tasks in enterprise software. This uncertainty shows up as non-linear navigation, backtracking, excessive pausing, and frequent help-documentation access. The approach described here quantifies this uncertainty using information-theoretic entropy measurement rather than analyzing what the employee says or how they score on assessments. Technical Description The system instruments enterprise UI workflows by injecting event listeners that capture a continuous stream of discrete UI events. Each event record contains: timestamp, action_type (click, hover, scroll, keypress, navigation, help_access), target_element_id, and session_id. Events are collected into per-session sequences grouped by workflow_type. For each workflow type, the system maintains an Optimal Workflow Model (OWM) derived from the navigation patterns of top-quartile performers. The OWM is represented as a first-order Markov chain where states are UI elements and transition probabilities are computed from expert user sessions. The model is rebuilt monthly from the trailing 90 days of top-performer data. For each employee session, the system computes Shannon Entropy over the observed navigation path: H(X) = negative sum of p(x_i) times log base 2 of p(x_i), where p(x_i) is the probability of each observed transition given the OWM. High entropy indicates the employee's navigation pattern deviates significantly from the expected expert pattern. The system also computes three supplementary signals: (a) backtrack_ratio, the count of backward navigation events divided by total navigation events; (b) dwell_anomaly_score, the count of UI elements where the employee's dwell time exceeded the OWM mean by more than 2 standard deviations; and (c) help_access_rate, the count of help documentation accesses per workflow instance. A Workflow Volatility Score V is computed as a weighted combination of the entropy and supplementary signals: V = w1 times H plus w2 times backtrack_ratio plus w3 times dwell_anomaly_score plus w4 times help_access_rate, with configurable weights (defaults: w1=0.40, w2=0.25, w3=0.20, w4=0.15). A dynamic threshold theta is computed for each workflow type as the mean V plus 1.5 times the standard deviation of V across the employee's peer group (same role, same workflow type) over the trailing 30-day window. When an employee's V exceeds theta for three or more consecutive workflow instances, the system generates a gap record containing: employee_id, workflow_type, volatility_scores, entropy_values, peer_threshold, triggering_session_ids, and the specific workflow stage where the highest entropy was observed. Distinguishing Characteristics This system does not process natural language communications. No NLP pipeline runs. No intent classification, entity extraction, or taxonomy mapping occurs. Gap detection operates on raw UI telemetry processed through information-theoretic entropy computation. The system also does not use structured assessment scores, quiz results, or Likert scales. The gap signal comes from how the employee navigates software, not from what they say or how they score.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Davis, Kenneth, "Shannon Entropy Workflow Analysis for Employee Knowledge Gap Inference", Technical Disclosure Commons, (March 25, 2026)
https://www.tdcommons.org/dpubs_series/9635