Identification of typing behaviors from large keystroke dataset

Abstract

In this thesis work, keystroke-level typing data of over 168000 participants are analyzed to understand determinants of transcription typing behaviors. Keystroke patterns are analyzed in detail and linked to typing performance. Inter-Key Intervals of letter pairs and other statistical indicators of typing performance are calculated and their distributions and statistical relations are studied. These analyses show, among other findings, that Inter-Key Intervals in typing distant letter pairs in the keyboard are more predictive than other letter pairs, e.g. letter repetitions. Rollover typing, where the next key is pressed before the previous key is released, is prevalent widely, linked to faster typing with high correlation. Finally, medoids-based (PAM) unsupervised clustering of participants is performed to identify groups of typists with similar typing characteristics, and the findings from the clusters are interpreted in terms of performance, accuracy, hand movements and rollover behaviors.