Page 4 - R_EdQuire White Paper Nov 2017 v3.4
P. 4
EdQuire White Paper: Computer learning behaviour in K-12
Nov 2017 V3.4
Page 4 of 15
Data collection
An edQuire software agent was installed on student computers, which securely transmitted student
computer usage data to our secure cloud service. Data connections were established with school time tabling
and student information systems. Data was only collected during lesson times and stored anonymously with
student identifying data held separately for security. Activities collected included applications, application’s
window title, visited websites and browser tab titles. Keystroke frequency and use of clipboard was recorded
(except in OS X). No typed content was recorded except for search engine text contained in URLs. Duration
spent on each website or application lasting > ½ a second was recorded. Data was automatically cleansed of
operating system activities. Activities without user input for 2 minutes were reclassified as Idle.
Data real-time display
Teachers were given an internet console for use in lessons, displaying the students‘ live on-taskness state
and history, allowing them to see at a glance classroom engagement and identify outlying, struggling or
unchallenged individual students who potentially needed support (Figure 1).
Figure 1. Teacher EdQuire classroom console. Horizontal bars at bottom of tiles represent a student’s history
in that lesson. Grey colour indicated Idle.
2.2 General Analysis method
Categorizing educategory - education vs entertainment for student computer activities
Student computer usage activities were first categorized in real time by a machine learning algorithm into
educational or non-educational. Activities were then automatically contextualized, according to teacher-
entered lesson keywords or subject topic, as:
• on teacher-assigned resource,
• On-Task (displayed green), on own discovered resource, OwnTask (blue), Off-Task (red) or
Unclassified/ambiguous (orange).
Ambiguous (orange) activities were resolved by opinions of educational experts and then fed back into the
algorithm for learning. The accuracy of our machine learning algorithm for educationally categorizing
activities was 93% when compared to human observers. For the purposes of this analysis, On-Task, OwnTask
and Ambiguous were grouped into On-Task.
© Copyright 2017 | All Rights Reserved by FIC Technology