CoChat
Sign Up
Cited by 66Personal Information Management and User BehaviorContext-Aware Activity Recognition SystemsHuman-Automation Interaction and Safety

Sensing Interruptibility in the Office: A Field Study on the Use of Biometric and Computer Interaction Sensors

Open access
Züger, Manuela, Müller, Sebastian C., Meyer, André N., Fritz, Thomas·2018·Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems
DOI 10.1145/3173574.3174165
Open this paper in CoChat — free accountOpen PDF

Abstract

Knowledge workers experience many interruptions during their work day. Especially when they happen at inopportune moments, interruptions can incur high costs, cause time loss and frustration. Knowing a person's interruptibility allows optimizing the timing of interruptions and minimize disruption. Recent advances in technology provide the opportunity to collect a wide variety of data on knowledge workers to predict interruptibility. While prior work predominantly examined interruptibility based on a single data type and in short lab studies, we conducted a two-week field study with 13 professional software developers to investigate a variety of computer interaction, heart-, sleep-, and physical activity-related data. Our analysis shows that computer interaction data is more accurate in predicting interruptibility at the computer than biometric data (74.8% vs. 68.3% accuracy), and that combining both yields the best results (75.7% accuracy). We discuss our findings and their practical applicability also in light of collected qualitative data.

Integrity signals

Registry

Crossref · 10.1145/3173574.3174165

Cites

92 of 92 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

5 of 60 cited sources checked · 2 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Prior research has linked several physiological signals-including heart rate, heart rate variability, electro-dermal activity, eye tracking, skin temperature, and EEG-to mental effort, cognitive load, task difficulty, emotions, and stress.

    “Various studies have shown that biometric data such as heart rate (HR), heart rate variability (HRV), electro-dermal activity (EDA), eye tracking, skin temperature or electroencephalography (EEG) can be used to assess mental effort and cognitive load [85,66,28,13], task difficulty [79,22], emotions [48,27,62], or stress [29,70,84].”
    Quote in the paperSubstantial claim1 of 9 sources checked
    • cites [85] 85. Glenn F Wilson. 2002. An analysis of mental workload in pilots during flight using multiple psychophysiological measures. The International Journal of Aviation Psychology 12, 1 (2002), 3-18. source
    • cites [66] 66. Peter Richter, Thomas Wagner, Ralf Heger, and Gunther Weise. 1998. Psychophysiological analysis of mental load during driving on rural roads-a quasi-experimental field study. Ergonomics 41, 5 (1998), 593-609. source
    • cites [28] 28. Eija Haapalainen, SeungJun Kim, Jodi F Forlizzi, and Anind K Dey. 2010. Psycho-physiological measures for assessing cognitive load. In Proceedings of the 12th ACM international conference on Ubiquitous computing. ACM, 301-310. source
    • cites [13] 13. Daniel Chen, Jamie Hart, and Roel Vertegaal. 2007. Towards a physiological model of user interruptability. In IFIP Conference on Human-Computer Interaction. Springer, 439-451. source
    • cites [79] 79. JA Veltman and AWK Gaillard. 1998. Physiological workload reactions to increasing levels of task difficulty. Ergonomics 41, 5 (1998), 656-669. source
    • cites [22] 22. Thomas Fritz, Andrew Begel, Sebastian C Müller, Serap Yigit-Elliott, and Manuela Züger. 2014. Using psycho-physiological measures to assess task difficulty in software development. In Proceedings of the 36th International Conference on Software Engineering. ACM, 402-413. source
    • cites [48] 48. Choubeila Maaoui, Alain Pruski, and F Abdat. 2010. Emotion recognition through physiological signals for human-machine communication. INTECH Open Access Publisher. source
    • cites [27] 27. Andreas Haag, Silke Goronzy, Peter Schaich, and Jason Williams. 2004. Emotion recognition using bio-sensors: First steps towards an automatic system. In Tutorial and research workshop on affective dialogue systems. Springer, 36-48. source
    • cites [62] 62. Sebastian C Müller and Thomas Fritz. 2015. Stuck and frustrated or in flow and happy: Sensing developers' emotions and progress. In Software Engineering (ICSE), 2015 IEEE/ACM 37th IEEE International Conference on, Vol. 1. IEEE, 688-699. source
  • Premisecitation

    Because the Fitbit samples at a coarser rate, it cannot measure heart rate variability.

    “While the Fitbit's coarser sampling granularity does not allow measuring HRV [83] and tends to overestimate sleep duration, it has a high intra-device reliability [60]”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [83] 83. Robert Wang, Gordon Blackburn, Milind Desai, Dermot Phelan, Lauren Gillinov, Penny Houghtaling, and Marc Gillinov. 2017. Accuracy of wrist-worn heart rate monitors. Jama cardiology 2, 1 (2017), 104-106. source

      The Methods passage (page not provided) describes recording Fitbit heart-rate readings at set points during a treadmill test. The Discussion passage (page not provided) says continuous assessment was not feasible with all devices, but neither passage identifies Fitbit’s sampling rate or addresses whether it can measure heart rate variability. Searches for HRV and sampling granularity found no passage establishing the claim. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
  • Findingown data

    Combining computer interaction and biometric data yields the best prediction accuracy for interruptibility (75.7%).

    “and that combining both yields the best results (75.7% accuracy)”
    • Also stated in: § INTRODUCTION · plain
      “while a combination of both was most accurate (75.7%)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported interruptibility ratings collected via experience sampling are a valid ground truth for predicting interruptibility.
    • Rests on an assumption: Participants complied with wearing the biometric sensors as instructed during the study.
    • Rests on an assumption: Participants provided accurate interruptibility ratings when prompted.
  • Findingown data

    A general classifier can achieve high accuracy (69.8%) for predicting interruptibility of new people.

    “a general classifier can achieve a high accuracy (69.8%)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported interruptibility ratings collected via experience sampling are a valid ground truth for predicting interruptibility.
    • Rests on an assumption: Participants complied with wearing the biometric sensors as instructed during the study.
    • Rests on an assumption: Participants provided accurate interruptibility ratings when prompted.
  • Findingown data

    An individual-specific classifier can beat the general model even when trained on only a small amount of data.

    “yet a classifier trained for a single individual can outperform the general one even with few data points”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported interruptibility ratings collected via experience sampling are a valid ground truth for predicting interruptibility.
    • Rests on an assumption: Participants complied with wearing the biometric sensors as instructed during the study.
    • Rests on an assumption: Participants provided accurate interruptibility ratings when prompted.
  • May combine two claims. 33
  • Classified as probably not a substantial claim. 28
  • The quote only partly states this claim. 3

  • May combine two claims.

    Premisecitation

    The Polar H7 uses an ECG-based sensor technique with medical grade accuracy.

    “The Polar H7 [17] is a chest strap recording heart beats and interbeat-intervals, using an electrocardiograph (ECG) based sensor technique with medical grade accuracy [83].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [83] 83. Robert Wang, Gordon Blackburn, Milind Desai, Dermot Phelan, Lauren Gillinov, Penny Houghtaling, and Marc Gillinov. 2017. Accuracy of wrist-worn heart rate monitors. Jama cardiology 2, 1 (2017), 104-106. source

      The Methods passage (page unavailable) identifies the Polar H7 as a chest-strap monitor, and the table reports 0.99 concordance with electrocardiogram-measured heart rate. The paper describes chest-strap monitors as electrode-based, but does not explicitly establish the Polar H7’s sensor technique or designate its accuracy as “medical grade.” (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Prior work indicates that sleep strongly affects productivity and mood.

    “A further and not yet fully studied factor of interruptibility is sleep, which has been shown to have a big impact on productivity and mood [68,80,49].”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [68] 68. Mark R Rosekind, Kevin B Gregory, Melissa M Mallis, Summer L Brandt, Brian Seal, and Debra Lerner. 2010. The cost of poor sleep: workplace productivity loss and associated costs. Journal of Occupational and Environmental Medicine 52, 1 (2010), 91-98. source
    • cites [80] 80. Stjepan Vidaček, Ljiljana Kaliterna, Biserka Radošević-Vidaček, and Simon Folkard. 1986. Productivity on a weekly rotating shift system: circadian adjustment and sleep deprivation effects? Ergonomics 29, 12 (1986), 1583-1590. source
    • cites [49] 49. Gloria Mark, Mary Czerwinski, Shamsi Iqbal, and Paul Johns. 2016. Workplace Indicators of Mood: Behavioral and Cognitive Correlates of Mood Among Information Workers. In Proceedings of the 6th International Conference on Digital Health Conference. ACM, 29-36. source

      Section 5.1 reports that more sleep was associated with more positive mood the next day. The Discussion (§6) suggests that poor sleep can lead to lower productivity and more negative mood, but presents this as a possible pathway requiring further study, not an established large causal effect on both outcomes. Page numbers are not available in our copy. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Interpretationcitation

    An interruptibility classifier can be used to show a knowledge worker's interruptibility to potential co-workers.

    “In general, such a classifier can then be used to indicate a knowledge worker's interruptibility to potential co-workers, which has been explored with physical indicators [87], or indicators displayed on the computer [9,46,21]).”
    Quote in the paperSubstantial claim1 of 4 sources checked
    • cites [87] 87. Manuela Züger, Christopher Corley, André N Meyer, Boyang Li, Thomas Fritz, David Shepherd, Vinay Augustine, Patrick Francis, Nicholas Kraft, and Will Snipes. 2017. Reducing Interruptions at Work: A Large-Scale Field Study of FlowLight. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems. ACM, 61-72. source
    • cites [9] 9. James Bo Begole, Nicholas E Matsakis, and John C Tang. 2004. Lilsys: sensing unavailability. In Proceedings of the 2004 ACM conference on Computer supported cooperative work. ACM, 511-514. source
    • cites [46] 46. Jennifer Lai, Sachiko Yoshihama, Thomas Bridgman, Mark Podlaseck, Paul B Chou, and Danny C Wong. 2003. MyTeam: Availability Awareness Through the Use of Sensor Data.. In INTERACT.
    • cites [21] 21. James Fogarty, Jennifer Lai, and Jim Christensen. 2004. Presence versus availability: the design and evaluation of a context-aware communication client. International Journal of Human-Computer Studies 61, 3 (2004), 299-317. source

  • May combine two claims.

    Findingown data

    The optimal time window for feature extraction varies by feature; user input features use 10-20 minutes and heart-related features use 2-3 minutes.

    “the optimal time windows vary per feature (e.g., 10-20min for user input and 2-3min for heart-related data)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported interruptibility ratings collected via experience sampling are a valid ground truth for predicting interruptibility.
    • Rests on an assumption: Participants complied with wearing the biometric sensors as instructed during the study.
    • Rests on an assumption: Participants provided accurate interruptibility ratings when prompted.

  • May combine two claims.

    Findingown data

    Participants reported that their interruptibility depends on internal states: sleepiness (85%), focus (77%), mood (46%), challenge (38%), productivity (38%), stress (38%), health (23%), and engagement (15%).

    “Participants further mentioned that their interruptibility depends on internal states such as sleepiness (85%), focus (77%), mood (46%), challenge (38%), productivity (38%), stress (38%), health (23%), and engagement (15%).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    All sensor sets performed better than the majority-class baseline, and computer interaction features were more predictive (74.8% accuracy) than biometric features (68.3% accuracy).

    “While all sensors were better than the baseline, the features of the computer interaction sensors (accuracy=74.8%)were more predictive compared to the features from the biometric sensors (accuracy=68.3%).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Most experience sampling prompts were answered quickly: half within 8 seconds and 83% within 15 minutes.

    “most prompts were answered quickly (50% within 8s, and 83% within 15 minutes).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    All interviewed participants said their interruptibility changes with computer activities such as coding or writing emails, but only 15% explicitly mentioned user input.

    “all participants stated in the interview that their interruptibility changes with certain activities on the computer, such as coding or writing emails, but only a few also explicitly mentioned the user input (15% of participants).”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    There was a non-significant trend that participants were more interruptible when sleepy (β=.11, p=.12).

    “and that there is a trend (not significant though) that participants were more interruptible when they were sleepy”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Calendar (2.8%), movement (2.3%), and time-related features (2.1%) were the least important, together contributing only 7.2%.

    “Calendar (2.8%), movement (2.3%) and time related features (2.1%) are the least important, contributing only 7.2%.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    When comparing the Polar and Fitbit sensors, Fitbit yielded better results for 9 of 13 participants, while Polar was more accurate for the remaining 4 (Fitbit accuracy 66.2%, Polar accuracy 62.5%).

    “When comparing the Polar and the Fitbit sensors, for 9 of 13 participants the Fitbit yielded better results, while for the remaining 4 the Polar was more accurate (accuracy Fitbit = 66.2%, accuracy Polar = 62.5%).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    One participant never used a rating of 1 or 2, producing a highly imbalanced two-state dataset (91.4% interruptible, 8.6% non-interruptible) under the 123|4567 split.

    “As one participant never used a rating of 1 or 2 and thus had a highly imbalanced dataset using this splitting method (for two states: 91.4% being interruptible -8.6% being non-interruptible)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For one participant (P06), the individual classifier began outperforming the general model with approximately 40 samples and improved as sample size increased.

    “The illustrated example shows that already with few samples (approximately 40 in this case), the individual classifier starts outperforming the general model and improves with increasing sample size.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Communication had an optimal time window of 3 hours and software development of 2 minutes.

    “Exceptions were communication (3h) and software development (2min).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Average prediction accuracies were 75.3% for two states, 65.5% for three states, and 42.5% for seven states, corresponding to improvements over a majority classifier of 26.6%, 25.7%, and 36.9%, respectively.

    “Average prediction accuracies were 75.3%, 65.5% and 42.5% for prediction into two, three and seven states of interruptibility, which is an average improvement of 26.6%, 25.7% and 36.9% compared to a majority classifier.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Experience sampling prompts were answered across the entire workday, with lower response rates early in the morning, at lunch, and in the evening.

    “Figure 2 shows that prompts were answered throughout the whole work day, though less often early in the morning, at lunch and in the evening”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Misclassifications in the three- and seven-state predictions rarely fell into distant classes and often fell into adjacent ones.

    “The aggregated confusion matrices for prediction into three and seven states reveal that mis-classifications rarely fall into distant classes, but often into adjacent ones (see Table 4 for seven states).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Individual models performed better than the general model for almost all participants except P03, P05, and P12, with an average accuracy of 75.3% versus 69.8% for the general model.

    “At the same time, and not surprising, the individual models performed better for almost all participants, except for three (P03, P05, and P12), with a 75.3% averaged accuracy over all participants compared to 69.8% for the general model.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For certain feature groups, longer time windows were more informative, and even daily features such as sleep had importance for predicting interruptibility (4.4% importance).

    “Our results show that for certain feature groups, longer time windows are more informative and that even daily features have an importance for predicting interruptibility, e.g.4.4% importance for sleep (see Table 2).”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Participants were less interruptible when they were engaged (β=-.41, p=.000).

    “and less when they were engaged”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    For 15 of the 55 features, multiple peaks or substantially different metric peaks led to adding more than one time window.

    “The latter occurred for 15 of the 55 features for which we determined a time window.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Participants perceived computer interaction sensors as less invasive but considered their captured data more sensitive than biometric data in the work context.

    “Overall, participants perceived the computer interaction sensors as less invasive, but thought that the captured data was more sensitive than the biometric data in the work context.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Many participants did not think there was a direct link between time of day and interruptibility, but instead linked it to activity routines and external factors such as background noise and interruption frequency.

    “many participants stated that they do not necessarily think that there is a direct link to interruptibility, but rather that the routine of activities and external factors such as background noise and interruption frequency is linked and might vary throughout the day.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The four univariate metrics were highly correlated and had similar peaks: Pearson's R with f_classif r=.92, with f_regression r=.95, and with mutual_info_regression r=.84.

    “We found that all four metrics were highly correlated, even the mutual information metric (Pearson's R and f_classif: Pearson r=.92, p<.000001, Pearson's R and f_regression: Pearson r=.95, p=.0, Pearson's R and mutual_info_regression: Pearson r=.84, p<.000001) and that they have similar peaks”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    In a multiple linear regression with 151 daily responses, participants were more interruptible when they had many interruptions (β=.14, p=.025).

    “The results (shown in Table 5) show that participants were more interruptible when they had many interruptions”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Computer interaction features such as keyboard/mouse input or application usage can be used to find suitable moments for interruptions.

    “Studies investigating computer interaction use features such as keyboard/mouse input or application usage to find suitable moments for interruptions [20,37].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [20] 20. James Fogarty, Andrew J Ko, Htet Htet Aung, Elspeth Golden, Karen P Tang, and Scott E Hudson. 2005. Examining task engagement in sensor-based statistical models of human interruptibility. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems. ACM, 331-340. source
    • cites [37] 37. Shamsi T Iqbal and Brian P Bailey. 2008. Effects of intelligent notification management on users and their tasks. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. ACM, 93-102. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Heart rate zones followed the American Heart Association definitions used by Fitbit: below 50% out of zone, 50-69% low, 70-84% high, and 85% or more peak.

    “We use heart rate zones as suggested by the American Heart Association and used by Fitbit: up to 49% of the maximum heart rate is regarded as being out of zone, 50% to 69% is labeled with low activity, 70% to 84% high activity and 85% and more is peak activity [4].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] 4. American Heart Association. 2016. Target Heart Rates. (Oct. 2016). http://www.heart.org/HEARTORG/ HealthyLiving/PhysicalActivity/FitnessBasics/ Target-Heart-Rates_UCM_434341_Article.jsp

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Computer interaction sensing covers only certain interactions and only daytime computer use, whereas biometric sensing is more invasive physically and more vulnerable to noise.

    “Especially since computer interaction sensors are limited to a specific kind of interaction and work during the day and biometric sensors are more physically invasive and more sensitive to noise (e.g.movement artifacts)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The study used a 7-point Likert scale for interruptibility ratings, extended from a 5-point scale because pilot participants tended to avoid extreme or intermediate parts of the scale.

    “In the pilot study, we further observed that some participants tended to avoid the extreme or intermediate parts of the scale.Therefore we extended the original 5-point Likert scale (which has predominantly been used in related work [20,76]) to a 7-point Likert scale to obtain a higher variety of ratings.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [20] 20. James Fogarty, Andrew J Ko, Htet Htet Aung, Elspeth Golden, Karen P Tang, and Scott E Hudson. 2005. Examining task engagement in sensor-based statistical models of human interruptibility. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems. ACM, 331-340. source
    • cites [76] 76. Takahiro Tanaka and Kinya Fujita. 2011. Study of user interruptibility estimation based on focused application switching. In Proceedings of the ACM 2011 conference on Computer supported cooperative work. ACM, 721-724. source

  • May combine two claims.

    Premisenone given

    Prior research on interruptibility predominantly used a single data type and short lab studies (less than three hours).

    “While prior work predominantly examined interruptibility based on a single data type and in short lab studies”
    • Also stated in: § INTRODUCTION · plain
      “the studies were predominantly conducted on small and controlled tasks over short periods of time (less than three hours) and mostly limited to either computer interaction or biometric sensors”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning

    The Fitbit tends to overestimate sleep duration.

    “While the Fitbit's coarser sampling granularity does not allow measuring HRV [83] and tends to overestimate sleep duration, it has a high intra-device reliability [60]”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Accelerometer measurements have been employed to identify physical activity and to indicate that activity transitions, such as walking between locations, are better times for interruptions.

    “Furthermore, accelerometer data has been used in several studies to detect physical activity and to show that interruptions are better delivered during moments recognized as activity transitions, e.g. when walking to another location [30,18,45].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [30] 30. Joyce Ho and Stephen S Intille. 2005. Using context-aware computing to reduce the perceived burden of interruptions from mobile devices. In Proceedings of the SIGCHI conference on Human factors in computing systems. ACM, 909-918. source
    • cites [18] 18. Robert Fisher and Reid Simmons. 2011. Smartphone interruptibility using density-weighted uncertainty sampling with reinforcement learning. In Machine Learning and Applications and Workshops (ICMLA), 2011 10th International Conference on, Vol. 1. IEEE, 436-441. source
    • cites [45] 45. Kyohei Komuro, Yuichiro Fujimoto, and Kinya Fujita. 2017. Relationship Between Worker Interruptibility and Work Transitions Detected by Smartphone. In International Conference on Human-Computer Interaction. Springer, 687-699. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The study extracted focus duration and activity/category switching frequency features inspired by Sarkar's work on mental fatigue of software developers.

    “We further extracted features related to focus duration and activity / category switching frequency inspired by Sarkar mental fatigue of software developers [71].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [71] 71. Saurabh Sarkar and Chris Parnin. 2017. Characterizing and predicting mental fatigue during programming tasks. In Proceedings of the 2nd International Workshop on Emotion Awareness in Software Engineering. IEEE Press, 32-37. source

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The study displayed interruptibility prompts at random intervals between 10 and 40 minutes.

    “The prompts asked participants to rate their current interruptibility on a 7-point Likert scale and were displayed in random intervals between 10 and 40 minutes.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    The study discarded data from one participant due to a technical issue with the Polar sensor, leaving 13 participants.

    “We discarded the data of one participant due to a technical issue with the Polar sensor that led to no recordings from this sensor and thus an incomplete and incomparable dataset for this participant.Of the remaining 13 participants, 1 was female and 12 were male.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The study used a pilot study with 8 software developers over 7 work days to inform the final procedure.

    “This decision was based on our experience from a pilot study with 8 software developers during 7 work days and from testing the final study procedure ourselves for several days.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The study used an automatic algorithm developed by Meyer et al. to map obvious programs and activities to activity categories.

    “First, an automatic algorithm developed by Meyer et al. mapped obvious programs and activities, such as Microsoft Visual Studio belonging to the activity category Coding [55].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55] 55. Andre N Meyer, Laura E Barton, Gail C Murphy, Thomas Zimmermann, and Thomas Fritz. 2017. The Work Life of Developers: Activities, Switches and Perceived Productivity. IEEE Transactions on Software Engineering (2017). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The study used common activity categories for software developers previously identified by Meyer et al.

    “We used common categories typical for software developers that had previously been identified by Meyer et al. [56].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] 56. André N Meyer, Thomas Fritz, Gail C Murphy, and Thomas Zimmermann. 2014. Software developers' perceptions of productivity. In Proceedings of the 22nd ACM SIGSOFT International Symposium on Foundations of Software Engineering. ACM, 19-29. source

  • May combine two claims.

    Interpretationown data

    The optimal time window varies by feature, and certain ranges of time windows work well for particular feature groups.

    “Our results show that the optimal time window varies per feature and suggest a range of time windows which work well for certain feature groups.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    The main advantage of a general model is that no initial training phase is needed to use it on a new subject in practice.

    “The main advantage of a general model is that no initial training phase is needed to use it on a new subject in practice.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    Learning curves for each individual were produced using shuffle split cross-validation with 100 splits and a test size of 20% of available samples.

    “we produced learning curves for each individual using shuffle split cross-validation (100 splits, test size of 20% of the available samples).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Previous research has linked both computer interaction and biometric sensors to mental load and interruptibility.

    “Previous research has already linked both, computer interaction and biometric sensors to mental load and interruptibility [28,45,80,37,63,20].”
    Quote in the paperSubstantial claim0 of 6 sources checked
    • cites [28] 28. Eija Haapalainen, SeungJun Kim, Jodi F Forlizzi, and Anind K Dey. 2010. Psycho-physiological measures for assessing cognitive load. In Proceedings of the 12th ACM international conference on Ubiquitous computing. ACM, 301-310. source
    • cites [45] 45. Kyohei Komuro, Yuichiro Fujimoto, and Kinya Fujita. 2017. Relationship Between Worker Interruptibility and Work Transitions Detected by Smartphone. In International Conference on Human-Computer Interaction. Springer, 687-699. source
    • cites [80] 80. Stjepan Vidaček, Ljiljana Kaliterna, Biserka Radošević-Vidaček, and Simon Folkard. 1986. Productivity on a weekly rotating shift system: circadian adjustment and sleep deprivation effects? Ergonomics 29, 12 (1986), 1583-1590. source
    • cites [37] 37. Shamsi T Iqbal and Brian P Bailey. 2008. Effects of intelligent notification management on users and their tasks. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. ACM, 93-102. source
    • cites [63] 63. Rahul Nair, Stephen Voida, and Elizabeth D Mynatt. 2005. Frequency-based detection of task switches. In Proceedings of the 19th British HCI Group Annual Conference, Vol. 2. 94-99.
    • cites [20] 20. James Fogarty, Andrew J Ko, Htet Htet Aung, Elspeth Golden, Karen P Tang, and Scott E Hudson. 2005. Examining task engagement in sensor-based statistical models of human interruptibility. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems. ACM, 331-340. source

  • Classified as probably not a substantial claim.

    Study design and conductown data

    Leave-one-out cross-validation was used to compare a general model trained on all participants except one against the held-out participant.

    “we used leave-one-out cross-validation for which we iterated over all participants and trained a classifier with data from all participants except one and tested it”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [81] 81. Aku Visuri, Niels van Berkel, Chu Luo, Jorge Goncalves, Denzil Ferreira, and Vassilis Kostakos. 2017. Predicting Interruptibility for Manual Data Collection: A Cluster-Based User Model. (2017). source

  • May combine two claims.

    Premisecitation

    Prior work has used various time windows for interruptibility prediction, mostly between 1 second and 5 minutes, but provides no general guideline for matching time windows to features.

    “While previous researchers have used a variety of time windows for predicting interruptibility, predominantly between 1s and 5mins [34,20,88], there is no general guideline on which time windows to use for which feature.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [34] 34. Scott Hudson, James Fogarty, Christopher Atkeson, Daniel Avrahami, Jodi Forlizzi, Sara Kiesler, Johnny Lee, and Jie Yang. 2003. Predicting human interruptibility with sensors: a Wizard of Oz feasibility study. In Proceedings of the SIGCHI conference on Human factors in computing systems. ACM, 257-264. source
    • cites [20] 20. James Fogarty, Andrew J Ko, Htet Htet Aung, Elspeth Golden, Karen P Tang, and Scott E Hudson. 2005. Examining task engagement in sensor-based statistical models of human interruptibility. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems. ACM, 331-340. source
    • cites [88] 88. Manuela Züger and Thomas Fritz. 2015. Interruptibility of software developers and its prediction using psycho-physiological sensors. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems. ACM, 2981-2990. source

  • May combine two claims.

    Interpretationreasoning

    The findings show that certain ranges of time windows exist for certain feature categories, but there is not necessarily a single best time window for each feature.

    “In general, our findings show that there are certain ranges of time windows for certain feature categories, but that there is not necessarily just one best time window for each feature.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Computer interaction features can accurately predict interruptibility at the computer and are more predictive than the biometric features used in this study, while biometric sensors still show strong potential for field prediction despite noise.

    “While our study demonstrates that computer interaction features can be used to accurately predict interruptibility at the computer and that they are more predictive than the biometric features used in our study, the results also show that biometric sensors already have a great potential in accurately predicting interruptibility in the field despite the noise.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Feature selection drew on prior work linking certain measures to cognitive load, stress, emotions, and interruptibility.

    “We extracted features that have previously been linked to cognitive states such as cognitive load, stress or emotions, and also interruptibility.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    A previous study found that office workers feel less productive after spending a longer amount of time on email activity, which may in turn affect their interruptibility.

    “This longer lasting effect of communication related activities was also mentioned in a previous study that found that office workers feel less productive after spending a longer amount of time with email activity [53], which in turn might impact their interruptibility.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [53] 53. Gloria Mark, Shamsi T Iqbal, Mary Czerwinski, Paul Johns, Akane Sano, and Yuliya Lutchyn. 2016. Email duration, batching and self-interruption: Patterns of email use on productivity and stress. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems. ACM, 1717-1728. source

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    For each participant, interruptibility was predicted using ten trials of stratified ten-fold cross-validation that preserved class proportions in each fold, with a random forest pipeline (500 estimators), initial feature imputation, and standard scaling.

    “For each participant, we predicted interruptibility using ten trials of stratified ten-fold cross-validation, which keeps the class proportions consistent in each fold, and a random forest classifier pipeline (500 estimators) with initial feature imputation and standard scaling.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Biometric sensors can serve as a complement or substitute to improve accuracy or to respect privacy preferences.

    “Biometric sensors can thus serve as a complement or substitute to improve accuracy, or respect privacy preferences for now.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Biometric and computer interaction data are highly individual, and trained models often cannot easily be transferred to new participants.

    “Since biometric and computer interaction data is highly individual and trained models can often not easily be transferred to new participants [22,88,81]”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [22] 22. Thomas Fritz, Andrew Begel, Sebastian C Müller, Serap Yigit-Elliott, and Manuela Züger. 2014. Using psycho-physiological measures to assess task difficulty in software development. In Proceedings of the 36th International Conference on Software Engineering. ACM, 402-413. source
    • cites [88] 88. Manuela Züger and Thomas Fritz. 2015. Interruptibility of software developers and its prediction using psycho-physiological sensors. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems. ACM, 2981-2990. source
    • cites [81] 81. Aku Visuri, Niels van Berkel, Chu Luo, Jorge Goncalves, Denzil Ferreira, and Vassilis Kostakos. 2017. Predicting Interruptibility for Manual Data Collection: A Cluster-Based User Model. (2017). source

  • May combine two claims.

    Interpretationown data

    A developer's interruptibility is affected not only by the seconds and minutes immediately before an interruption, but also by features such as activities or sleep that can have longer-lasting effects.

    “Our results suggest that a developer's interruptibility is not only affected by the few seconds and minutes before an interruption, but that there are features, such as the activities or sleep, that can have a longer lasting effect on interruptibility.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    An interruptibility classifier can be used to mediate interruptions by postponing computer-based interruptions until a person is more interruptible.

    “Similarly, such an interruptibility classifier can be used to mediate interruptions directly by postponing computer-based interruptions while a person is non-interruptible to a more opportune moment, which has also been investigated in prior work [37,32,41,42,33].”
    Quote in the paperNot a substantial claim0 of 5 sources checked
    • cites [37] 37. Shamsi T Iqbal and Brian P Bailey. 2008. Effects of intelligent notification management on users and their tasks. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. ACM, 93-102. source
    • cites [32] 32. Eric Horvitz, Paul Koch, and Johnson Apacible. 2004. BusyBody: creating and fielding personalized models of the cost of interruption. In Proceedings of the 2004 ACM conference on Computer supported cooperative work. ACM, 507-510. source
    • cites [41] 41. Ashish Kapoor and Eric Horvitz. 2007. Principles of lifelong learning for predictive user modeling. User Modeling 2007 (2007), 37-46. source
    • cites [42] 42. Ashish Kapoor and Eric Horvitz. 2008. Experience sampling for building predictive user models: a comparative study. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. ACM, 657-666. source
    • cites [33] 33. Eric Horvitz, Paul Koch, Carl M Kadie, and Andy Jacobs. 2002. Coordinate: Probabilistic forecasting of presence and availability. In Proceedings of the Eighteenth conference on Uncertainty in artificial intelligence. Morgan Kaufmann Publishers Inc., 224-233. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Data on computer interaction is inherently restricted to periods when the person is at the computer.

    “e.g., computer interaction data is inherently limited to times spent at the computer.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Missing values were imputed by replacing them with the mean before classification.

    “Missing values were imputed by replacing with the mean before classification”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study spent enough time explaining the pop-ups at the beginning to mitigate the risk that participants might not assess their interruptibility correctly or understand the question.

    “We ensured that we spent enough time to explain the pop-ups at the beginning of the study to mitigate this risk.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    Imputing missing values with the mean can yield better results than discarding them, which would reduce the sample size.

    “as this technique can lead to better results than discarding them which would decrease the sample size.”
    Quote in the paperNot a substantial claim
57 of 57

  • The authors state this as an assumption.

    Physiological signals such as heart rate, pupil dilation, or brain activity are linked to cognitive states and task engagement and can therefore indicate interruptibility.

    “Studies on biometric sensors are based on the assumption that physiological features, such as heart rate, pupil dilation or brain activity, can be linked to the user's cognitive states and task engagement and thus be used to determine interruptibility [14,6,88].”

  • The paper relies on this without support.

    Self-reported interruptibility ratings collected via experience sampling are a valid ground truth for predicting interruptibility.

    “we collected interruptibility ratings through experience sampling using a pop-up displayed on the computer, that we then used as ground truth for predicting a participant's interruptibility.”
    “Interruptibility Ratings.To collect the ground truth for the interruptibility classification, we prompted our participants with an experience sampling technique using a pop-up that was displayed on the computer.”

  • The paper relies on this without support.

    Participants complied with wearing the biometric sensors as instructed during the study.

    “We asked the participants to wear the Polar sensor during work hours, and the Fitbit sensor as much as possible including work and free time as well as nights, except when they did not feel comfortable wearing it or when swimming, showering or charging the device.”

  • The paper relies on this without support.

    Participants provided accurate interruptibility ratings when prompted.

    “The prompts asked participants to rate their current interruptibility on a 7-point Likert scale and were displayed in random intervals between 10 and 40 minutes.”

  • The paper relies on this without support.

    The monitoring tool captured all relevant computer interaction data from participants' work computers.

    “In addition, we installed a monitoring tool to collect computer interaction data.In case a participant worked on several computers, we installed the monitoring tool on all of them to collect a complete data set.”

  • The authors state this as an assumption.

    This study is the first to compare computer interaction and biometric sensors for interruptibility in the field over a longer period.

    “To the best of our knowledge, our study is the first to compare these types of sensors in the field over a longer period of time.”

  • The authors state this as an assumption.

    Findings can be extended to comparable job roles and settings.

    “While we can assume that the results can be generalized to similar job roles and environments”
7 of 7

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Automated task switch detection could improve interruptibility classifiers and postpone in-person interruptions from co-workers to task switch borders, times they are less costly.

    “Similarly, an automated task switch detection could improve interruptibility classifiers and postpone in-person interruptions from co-workers to task switch borders, times they are less costly [70], [71], [72].”

    From Detecting Developers' Task Switches and Types (2022) · cites this paper as [72]

    In the Discussion (page unavailable), the paper suggests that a short-window computer-interaction feature might indicate a breakpoint or task switch. In “Finding Opportune Moments for Interruptions” (page unavailable), it describes task boundaries as opportune moments and says interruptibility estimates can help co-workers avoid in-person interruptions at inopportune times. It does not say that automated task-switch detection would improve classifier performance or specifically postpone in-person interruptions until task switches. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Not confirmedF(8, 142) = 8.34reported p < 0 · recomputed p < .001

    “F(8, 142)=8.34,p < .0”

1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Self-check results appear here once an independent check has confirmed them.

Cites

Flagged first, then by citations

Research works printed but not linked: 8

  1. 10.5555/2042118.2042162 · DOI
  2. American Heart Association. 2016. Target Heart Rates. (Oct. 2016). http://www.heart.org/HEARTORG/ HealthyLiving/PhysicalActivity/FitnessBasics/ Target-Heart-Rates_UCM_434341_Article.jsp
  3. 10.5555/1778331.1778380 · DOI
  4. Polar Electro. 2017. Equine H7 heart rate sensor belt set. https: //www.polar.com/en/products/equine/accessories/ equine_H7_heart_rate_sensor_belt_set. (2017). {Online; accessed 19-September-2017}.
  5. Inc. Fitbit. 2017. Fitbit Charge 2. https://www.fitbit.com/de/charge2. (2017). {Online; accessed 19-September-2017}.
  6. Microsoft. 2017. Microsoft Graph. https://graph.microsoft.io. (2017). {Online; accessed 19-September-2017}.
  7. 10.5555/2011878 · DOI
  8. 10.5555/3105556.3105566 · DOI

Cited by

Sorted by citations

Related

Authors
Züger, M., Müller, S. C., Meyer, A. N., & Fritz, T.
Affiliations
  • University of Zurich (CH)
  • University of Zurich, Zürich, Switzerland
ORCID
Venue
Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2018-04-21
Pages
1-14
DOI
10.1145/3173574.3174165
OpenAlex
W2796288789
Cited by
66
References
87
Citations by year
2026: 8 · 2025: 13 · 2024: 8 · 2023: 9 · 2022: 2 · 2021: 6 · 2020: 12 · 2019: 6 · 2018: 2
FWCI
15.69
Type
conference paper
Language
en
Open access
Licence
Locations
  • CHI Conference on Human Factors in Computing Systems (publishedVersion, other-oa)
  • Zurich Open Repository and Archive (University of Zurich) (acceptedVersion, other-oa)
  • Zurich Open Repository and Archive (University of Zurich) (submittedVersion, other-oa)
  • Zurich Open Repository and Archive (University of Zurich)
Funders
SNF (200021_150050)
Topics
  • Social Sciences › Decision Sciences › Information Systems and Management › Personal Information Management and User Behavior
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Social Sciences › Psychology › Social Psychology › Human-Automation Interaction and Safety
Keywords
interruptibility, knowledge workers, interruptions, software developers, biometric data, heart rate, sleep data, physical activity, office work, multimodal fusion

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-09-29.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-29.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review