# How We Swipe: The Dataset The `swipelogs.zip` file comprises 1338 `.log` (swipe data) and associated `.json` (metadata) files. The uncompressed data takes about 1 GB of disk space. **Note:** Our typing study was available "in the wild", so a few logs are ill-formed and even contain non-English words. We recommend to use the log parsers provided in https://github.com/luileito/swipetest/tree/master/experiments ## Swipe log file example This is an excerpt of `g1dgig1st0he01bmgdjo6crgds.log` file (space-separated format): ```csv sentence timestamp keyb_width keyb_height event x_pos y_pos x_radius y_radius angle word is_err force_instrumental_reliance_replacement 1576480557856 360 205 touchstart 140 100 1.3324081897736 1.3324081897736 0 force 0 force_instrumental_reliance_replacement 1576480557864 360 205 touchmove 140 100 1.4989590644836 1.4989590644836 0 force 0 force_instrumental_reliance_replacement 1576480557879 360 205 touchmove 140 100 1.1658570766449 1.1658570766449 0 force 0 force_instrumental_reliance_replacement 1576480557898 360 205 touchmove 142 99 1.1658570766449 1.1658570766449 0 force 0 force_instrumental_reliance_replacement 1576480557920 360 205 touchmove 167 90 1.3324081897736 1.3324081897736 0 force 0 force_instrumental_reliance_replacement 1576480557945 360 205 touchmove 232 67 1.4989590644836 1.4989590644836 0 force 0 force_instrumental_reliance_replacement 1576480557963 360 205 touchmove 253 60 1.1658570766449 1.1658570766449 0 force 0 force_instrumental_reliance_replacement 1576480557982 360 205 touchmove 266 56 0.99930608272552 0.99930608272552 0 force 0 force_instrumental_reliance_replacement 1576480557991 360 205 touchmove 270 54 1.3324081897736 1.3324081897736 0 force 0 force_instrumental_reliance_replacement 1576480558007 360 205 touchmove 286 48 1.4989590644836 1.4989590644836 0 force 0 ``` The columns comprise the following information: * `sentence` (string) Prompted sentence. Words are separated by underscores. * `timestamp` (integer) Time (in milliseconds) at which the event was created. * `keyb_width` (integer) Width of the rendered virtual keyboard. * `keyb_height` (integer) Height of the rendered virtual keyboard. * `event` (string) Name of the event. * `x_pos` (integer) X coordinate of the touch point relative to the keyboard width. * `y_pos` (integer) Y coordinate of the touch point relative to the keyboard height. * `x_radius` (float) X radius of the ellipse that most closely circumscribes the touch contact area. * `y_radius` (float) Y radius of the ellipse that most closely circumscribes the touch contact area. * `angle` (float) Rotation angle, in degrees, of the contact area ellipse defined by `x_radius` and `x_radius`. * `word` (string) Swiped word. * `is_err` (bool) Flag indicating whether the swiped word was entered correctly (0) or not (1). More info about touch events: https://developer.mozilla.org/en-US/docs/Web/API/Touch_events ## User metadata file example This is the content of `g1dgig1st0he01bmgdjo6crgds.json` file: ```json {"gender":"Male","age":40,"nationality":"ES","familiarity":"Everyday","englishLevel":"Advanced","dominantHand":"Right","swipeHand":"Right","swipeFinger":"Index","screenWidth":360,"screenHeight":720,"devicePixelRatio":3,"maxTouchPoints":5,"platform":"Linux armv8l","vendor":"Google Inc.","referal":"","language":"en-US,en;q=0.9","userAgent":"Mozilla\/5.0 (Linux; Android 7.1.1; K10) AppleWebKit\/537.36 (KHTML, like Gecko) Chrome\/79.0.3945.79 Mobile Safari\/537.36","timestamp":1576480549} ``` ## Processed files We provide several processed files to ease further analyses. These processed files refer to the subset of 909 users that we reported in our paper. The procedure we performed is the following: - Select users who submitted 5 or more sentences (one third of the requested sentences). - Select mobile users (`maxTouchPoints > 2`, screen size criteria: `width < 600` and `500 < height < 900`). - Remove users having `NA`s in their demographic information. ### Metadata Log metadata, such as user demographics, language, device platform, etc. are provided in `metadata.tsv` file (tab-separated format). ```csv age dominant_hand english_level familiarity gender language max_touch_points nationality screen_height screen_width swipe_finger swipe_hand timestamp uid vendor 19 right beginner sometimes female en 5 ae 640 360 thumb right 1583309231 0q1sat7ola2osk0ohdtciaboba google inc. 19 left native rarely female en 5 us 869 412 index right 1583136907 124p7a10vadpv7d0rk37oiv3g5 google inc. 44 left beginner rarely female en 5 mx 640 360 index both 1582648583 19hubm5jt63707ejbcnkimlr08 google inc. 18 right intermediate never female mt 5 mt 892 412 thumb right 1582555916 20sa5uaiit0t55or9miatm70tv google inc. 38 right advanced rarely male es 5 mx 846 412 thumb both 1582073116 2r7mdfhcjesci8trr5afg2akq5 google inc. 18 right advanced sometimes male en 5 ca 667 375 thumb right 1581726141 2sd22gsn7kdnhs3mrmh93j7hvk apple computer, inc. 25 right native rarely female en 5 us 896 414 other right 1582827386 3a064eetjpjpr55674f3qc5t55 apple computer, inc. 19 left native rarely male en 5 us 869 412 thumb right 1582557536 3fmgdlooume5t7fu6dula4rn8m google inc. 22 right native often female en 5 us 896 414 thumb right 1583387995 4ikrjsdgdlgfhvsbekfpo8605g apple computer, inc. ``` The file has 15 columns: * `uid` (string) User ID. It unequivocally links all dataset files, e.g. swipe logs, metadata, etc. * `timestamp` (integer) Time (in seconds) at which the user entered the study. * `nationality` (string) Self-reported user nationality, according to the ISO 3166-1 alpha-2 standard: https://en.wikipedia.org/wiki/ISO_3166-1_alpha-2 * `age` (integer) Self-reported user age. Possible values are between 18 and 99. * `gender` (string) Self-reported user gender. Possible values are: "male", "female", "other". * `dominant_hand` (string) Self-reported dominant hand. Possible values are: "left", "right". * `swipe_hand` (string) Self-reported input hand. Possible values are: "left", "right", "both". * `swipe_finger` (string) Self-reported input finger. Possible values are: "index", "thumb", "other". * `english_level` (string) Self-reported user gender. Possible values are: "beginner", "intermediate", "advanced", "native". * `familiarity` (string) Self-reported familiarity with shape-writing input. Possible values are: "never", "rarely", "sometimes", "often", "everyday". * `language` (string) Preferred user language, according to the HTTP Accept-Language header: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Accept-Language * `max_touch_points` (integer) Maximum number of simultaneous touch contact points supported by the device. * `screen_height` (integer) Height of the device screen, in CSS pixels. * `screen_width` (integer) Width of the device screen, in CSS pixels. * `vendor` (string) Navigator vendor property. Possible values are "google inc." or "apple computer, inc.". ### Sentence-level performance stats This is an excerpt of `stats-sentences.tsv` file (tab-separated format): ```csv good_wpm_swipe sentence fail_wpm_classic fail_wpm_swipe fail_interval_time fail_length fail_time good_length fail_dtw good_interval_time good_dtw good_time good_wpm_classic wer username dataset 18.184573420215184 they_have_capacity_now 15.933609958506224 9.95850622406639 nan 1409.5558752582206 3012.5 904.556376919928 10498.246089818047 1703.5 3660.1316217387975 1648.0 20.0030307622367 0.25 0q1sat7ola2osk0ohdtciaboba enron 39.267015706806276 what_a_jerk nan nan 453.5682207227957 nan nan 2146.824630286267 1528.0 35.3403141361256560.0 0q1sat7ola2osk0ohdtciaboba enron 23.170496234794363 what_is_the_mood nan nan 284.2313386802028 nan 1458.0 1512.468340490222 890.5 18.53639698783549 0.0 0q1sat7ola2osk0ohdtciaboba enron 15.671251958906494 im_on_a_plane 19.524894240156197 19.5248942401562 nan nan 103.51568765980456 nan 3994.5 452.42589317380464442.0 11.49225143653143 0.25 0q1sat7ola2osk0ohdtciaboba enron 12.061513719971856 mcdonald_click_adventures_amenities nan nan 396.0785894109886 nan 1758.0 5906.483283114141 2984.5 21.10764900995075 0.0 0q1sat7ola2osk0ohdtciaboba rand 22.253129346314328 we_are_all_fragile nan nan 218.1191927999088 nan 1478.0 1485.3408015189236 669.0 20.027816411682892 0.0 0q1sat7ola2osk0ohdtciaboba enron 8.916960802526471 mathematics_komatsu_mold_shop nan nan 712.6508009226677 nan 2663.0 4821.69478236841 3218.5 12.929593163663384 0.0 0q1sat7ola2osk0ohdtciaboba rand 14.232624337787616 arms_sichuan_asks_blogs nan nan 589.5511226552718 nan nan 5399.855165735805 3772.0 17.07914920534514 0.0 0q1sat7ola2osk0ohdtciaboba rand 4.908644668666485 generates_kentucky_holidays_etrade nan nan 947.2273284276482 nan 5530.5 13442.322231641021 10055.0 8.835560403599674 0.0 0q1sat7ola2osk0ohdtciaboba rand ``` The file has 16 columns: * `username` (string) User ID. * `sentence` (string) Prompted sentence. Words are separated by underscores. * `dataset` (string) Dataset which the sentence belongs to. Possible values are: "enron" (memorable sentences), "rand" (randomly generated sentences). * `good_length` (float) Median swipe length for the _successfully entered_ words. * `good_time` (float) Median swipe time for the _successfully entered_ words. * `good_interval_time` (float) Median interval time for the _successfully entered_ words. * `good_dtw` (float) Median swipe error of _successfully entered_ words, in pixels. It is computed as the elastic matching score against the ideal swipe trajectory, according to the Dynamic Time Warping (DTW) algorithm. * `good_wpm_classic` (float) Words per minute according to http://www.yorku.ca/mack/RN-TextEntrySpeed.html for the _successfully entered_ words. This is the WPM we report in our paper. * `good_wpm_swipe` (float) Words per minute computed by dividing the number of entered words by the overall time (including both swipe times and interval times) for the _successfully entered_ words. * `wer` (float) Median "a priori" word error rate, computed as the number of (unique) failed words divided by the number of reference words. This value is not reliable because users had to enter all words successfully in order to advance to the next sentence. * `fail_length` (float) Median swipe length for the _failed_ words. * `fail_time` (float) Median swipe time for the _failed_ words. * `fail_interval_time` (float) Median interval time for the _failed_ words. * `fail_dtw` (float) Median swipe error of _failed_ words, in pixels. It is computed as the elastic matching score against the ideal swipe trajectory, according to the Dynamic Time Warping (DTW) algorithm. * `fail_wpm_classic` (float) Words per minute, according to http://www.yorku.ca/mack/RN-TextEntrySpeed.html for the _failed_ words. * `fail_wpm_swipe` (float) Words per minute computed by dividing the number of entered words by the overall time (including both swipe times and interval times) for the _failed_ words. ### Word-level performance stats This is an excerpt of `stats-words.tsv` file (tab-separated format): ```csv dataset dtw is_failed length sentence time username word enron 4853.106069030182 0 1078.9805063431345 they_have_capacity_now 1709 0q1sat7ola2osk0ohdtciaboba they enron 2467.1571744474127 0 730.1322474967213 they_have_capacity_now 1587 0q1sat7ola2osk0ohdtciaboba have enron 17806.80973715143 0 1685.6432266743946 they_have_capacity_now 5574 0q1sat7ola2osk0ohdtciaboba capacity enron 2142.9952767392892 0 333.4053003465969 they_have_capacity_now 921 0q1sat7ola2osk0ohdtciaboba now enron 18134.877696786567 1 2381.1457347329333 they_have_capacity_now 5001 0q1sat7ola2osk0ohdtciaboba capacity enron 2861.6144828495258 1 437.9660157835077 they_have_capacity_now 1024 0q1sat7ola2osk0ohdtciaboba capacity enron 2116.483479831384 0 506.3568403260237 what_a_jerk 1399 0q1sat7ola2osk0ohdtciaboba what enron 2177.1657807411507 0 400.77960111956764 what_a_jerk 1657 0q1sat7ola2osk0ohdtciaboba jerk enron 1854.654769488807 0 495.8353143406047 what_is_the_mood 1176 0q1sat7ola2osk0ohdtciaboba what ``` The file has 8 columns: * `username` (string) User ID. * `dataset` (string) Dataset which the word belongs to. Possible values are: "rand2k" (highly frequent words), "rand3k" (common words), "rand5k" (infrequent words), "rand0" (out-of-vocabulary words). * `word` (string) Swiped word. * `is_failed` (bool) Whether the word was swiped successfully (0) or not (1). * `length` (float) Swipe length, in pixels. * `time` (integer) Swipe time, in milliseconds. * `dtw` (float) Swipe error, in pixels. It is computed as the elastic matching score against the ideal swipe trajectory, according to the Dynamic Time Warping (DTW) algorithm. ## More resources See https://github.com/luileito/swipetest for the source code of our online study, the phrase sets, and the processing scripts we used in our experiments. # Citation Please cite us using the following reference: - L. A. Leiva, S. Kim, W. Cui, X. Bi, A. Oulasvirta. **How We Swipe: A Large-scale Shape-writing Dataset and Empirical Findings.** *Proc. MobileHCI, 2021.* ```bib @InProceedigs{swipe_dataset, author = {Luis A. Leiva and Sunjun Kim Wenzhe Cui and Xiaojun Bi and Antti Oulasvirta}, title = {How We Swipe: A Large-scale Shape-writing Dataset and Empirical Findings}, booktitle = {Proc. MobileHCI}, year = {2021}, } ```