Process the 136M Keystrokes dataset to analyze bigram and word typing speeds.
Repository: https://github.com/binarybottle/process_3.5M_keystrokes.git
Author: Arno Klein (arnoklein.info)
License: MIT License (136M Keystrokes datasetsee LICENSE)
The contains keystroke data of over 168000 users typing 15 sentences each. The data was collected via an online typing test published at a free typing speed assessment webpage. https://userinterfaces.aalto.fi/136Mkeystrokes/ More details about the study and its procedure can be found in the paper:
Vivek Dhakal, Anna Maria Feit, Per Ola Kristensson, Antti Oulasvirta. Observations on Typing from 136 Million Keystrokes. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, ACM, 2018.
filter_participants.py filters the metadata_participants.txt file.
Criterion Breakdown (Total participants: 168594):
- LAYOUT = qwerty: 165324 participants (98.1%)
- FINGERS = 9-10: 107063 participants (63.5%)
- KEYBOARD_TYPE = full/laptop: 165009 participants (97.9%)
- ERROR_RATE < 1.167: 103495 participants (61.4%)
- ALL criteria combined: 69062 participants (41.0%)
Filtered data written to 'filtered_metadata_participants.txt'
Age Distribution: Mean: 25.9 years Range: 0 - 120 years Std Dev: 11.1
Gender Distribution: female: 33876 (49.1%) male: 28312 (41.0%) none: 6874 (10.0%)
Typing Course Training: Trained: 27747 (40.2%) Untrained: 41315 (59.8%)
Country Distribution: US: 47754 (69.1%) PH: 3823 (5.5%) CA: 3507 (5.1%) IN: 3238 (4.7%) GB: 1870 (2.7%) ... and 189 other countries
Native Language Distribution: en: 59112 (85.6%) tl: 1743 (2.5%) zh: 1422 (2.1%) es: 1100 (1.6%) hi: 589 (0.9%) ... and 139 other languages
Time Spent Typing Distribution: 0 hours/day: 4986 (7.2%) 1 hours/day: 20814 (30.1%) 2 hours/day: 11768 (17.0%) 3 hours/day: 6396 (9.3%) 4 hours/day: 5226 (7.6%) 5 hours/day: 5176 (7.5%) 6 hours/day: 3685 (5.3%) 7 hours/day: 1375 (2.0%) 8 hours/day: 5790 (8.4%) 9 hours/day: 661 (1.0%) 10 hours/day: 1462 (2.1%) 11 hours/day: 35 (0.1%) 12 hours/day: 671 (1.0%) ...
Typing Speed (WPM) Analysis: Mean: 58.1 WPM Range: 6.0 - 152.6 WPM Std Dev: 20.0
process_keystroke_data.py processes keystroke files for filtered participants to calculate:
- Interkey intervals for correctly typed letter bigrams (excludes spaces, Shift, etc.)
- Word timing data (first to last letter press time per word) for correctly typed multi-letter words
Only includes data for keystrokes that match the expected target text (no typing errors). Single-letter words are excluded from word timing analysis since they have no interkey intervals. Bigrams and words with intervals <30ms or >3000ms are excluded to filter out hardware issues and pauses.
Processed 61603/69062 participants Total correct bigrams collected: 19809877 Total correct multi-letter word times collected: 6158104 Correct bigram results (30-3000ms) written to: bigram_times.csv Correct multi-letter word timing results (30-3000ms) written to: word_times.csv
Correct Bigram Statistics (30-3000ms): Unique bigrams: 413 Mean interval: 172.2 ms Min interval: 30 ms Max interval: 3000 ms Most common bigrams: th: 708088 occurrences he: 589922 occurrences in: 504056 occurrences re: 376910 occurrences ou: 366279 occurrences