PowerPoint Presentation

Hooked on Smartphones: An Exploratory Study on Smartphone Overuse among College StudentsUichin Lee, Joonwon Lee, Minsam Ko, Subin Yang, Gahgene Gweon (KAIST, Knowledge Service Engineering) Changhun Lee, Yuhwan Kim, Junehwa Song (KAIST, Computer Science) Koji Yatani (Microsoft Research Asia)Kyong Mee Chung (Yeonsei University, Dept. of Psychology)

Thanks for the introduction; in this talk, we consider over-attachment on mobiles. this is a joint work w/ GG, June at KAIST; Koji at MSRA, and Kyong Mee from Dept. of Psychology at Yeonsei Univ. 1Smartphone Overuse

Smartphone usage have both positive and negative impacts on our lives. In our work, we wanted to talk about the negative aspects. First of all (CLICK), it is not very difficult to observe people on the street, typing and giggling while they are walking.And also (CLICK), when we hang out with friends, we use smartphones too often. When it comes to family lives (CLICK), we see this kind of situation quite often. (CLICK) And we have to admit that we use smartphones when we go to bed, and after we get up. Besides disruption of social interactions, there is also a growing concern with poor mental health such as attention deficits and sleep deprivation.

2Smartphone OveruseExcessive smartphone use can be pathological and addictive

Technological addiction: non-chemical (behavioral) addiction (Griffiths, 1995)

Human-machine interaction contains inducing and reinforcing features that promote addictive behaviors

(CLICK) Excessive smartphone use can sometimes be pathological and addictive as you can see in this picture. (CLICK) This kind of problematic behavior can be considered as technological addiction.Technological addiction is non-chemical (or behavioral) addiction.(CLICK) It is mainly due to the fact that human-machine interaction contains inducing and reinforcing features that promote addictive behaviors.

http://mobilefiller.com/2011/08/04/a-cure-for-smartphone-addiction/ http://www.hourdetroit.com/Hour-Detroit/January-2012/Tethered-by-Tech/

3Related Work & MotivationPsychology work mostly studies scale development, and psychological factors (Kim 2012, Carbonell 2012) Usage measurement work analyzes general usage patterns (Falaki 2010, Bhmer 2011)Recent HCI studies on smartphone overuse: College students practices of managing overuse (Ames 2013) Habits of update checking (emails, friend availability sharing) (Oulasvirta 2012)Yet, lack of understanding on detailed usage behaviors related to problematic usage of smartphonesRegarding mobile overuse, previous psychology studies mostly focused on developing measurement scales, and also finding psychological factors related to problematic usage. Since these studies rely on self-report data, only minimal usage data are considered for the analysis due to difficulties of usage recall . Usage measurement studies of smartphones, which could be useful to bridge this gap, so far mostly focused on analyzing general usage patterns of mobile devices, not really dealing with problematic usage.

And recently there is a growing interest in HCI communities on mobile overuse;Ames conducted an observational study on college students practice of overuse concerns in Stanford Campus- empirically showed that mobile devices induce habitual checking behavior like emails.

However, there is still lack of understanding on detailed usage behaviors related to problematic usage. 4Research OverviewIdentify detailed usage behaviors related to problematic usage of smartphones Self-report AddictionQuestionnaire

Real Usage Data Collection

Exit Interviews(usage patterns)

ContentAnalysisAnalysis of Between-groupUsage Differences

RiskNon-riskQuantitative analysisContent analysisTo bridge this gap, we use a mixed approach by performing both quantitative analysis of fine-grained usage data; and qualitative analysis of usage behavior interview data. (CLICK) First, we recruit people and ask them to answer self-report addiction questionnaire. (CLICK) We then install our smartphone usage logging software to collect fine-grained usage data. (CLICK) After data collection, we divide users into two groups based on each persons addiction score and explore any usage differences. (CLICK) We also performed exit interviews to supplement our usage data analysis (CLICK) so that we can have a deeper understanding of usage behaviors.

5Methodology: Participants95 college students in a large univ. (Fall, 2012)(67 males; and 28 females)

Avg. age of participants: 20.6 (SD 1.7)

Avg. participation duration: 26.8 days (SD 9.5)(more than 60,000 hours of usage data)

We hired 95 undergrad students in a large university in the fall semester, 2012.The average age of the participants were about 20. The average participation duration is about 27 days; and total hours of usage are more than 60,000.

6Methodology: Self-Report DataSmartphone addiction scale (Kim et al., 2012)

InterferenceMy school grades (or work productivity) dropped due to excessive smartphone use.Virtual worldorientationUsing a smartphone is more enjoyable than spending time with my family or friends.WithdrawalIt would be distressing if I am not allowed to use my smartphone.ToleranceEven when I think I should stop, I continue to use my smartphone.(15 items, based on well-known DSM-IVs addiction factors)Non-Risk group # users = 59Risk group(# users = 36)

To assess the degree of smartphone addiction, we used a smartphone addiction scale proposed by Kim et al. in 2012. It is a 15 item scale with four sub-domains: >> interference measures smartphones interference in daily activities; like my.. >> virtual world orientation has items like .. >> withdrawal of smartphones has items like: >> tolerance of usage has items like

(CLICK) The scale allows us to divide our participants into two groups: we had 36 users who belong to the risk group; and we had the rest of the participants in the non-risk group. 7

Methodology: Usage LoggingDeveloped SmartLogger, an unobtrusive logging tool, leveraging Androids accessibility service

Logging fine-grained usage data:System events (power on/off, screen on/off/unlock, battery status changes)App events (active/inactive apps; touch/text input events; web browsing URLs; notifications)Telephone events (call/SMS, ringer mode changes)

To collect the real usage data, we developed an unobtrusive logging tool for android It logs fine-grained usage data such as screen on/off, active/inactive apps, touch events, notifications, and telephone events 8Methodology: Usage AnalysisScreen OnScreen OffUnlockApp sequenceNotificationsapp1app1Usage Sessions(unlocked usage)Sessioni-1timeSessioni+1Sessioni

app1app2launcher

Inter-notification timeInter-sessiontimeSessiontimeHere is the user behavior model used for analysis. (CLICK) We see that this user had three usage sessions over period of time.(CLICK) this session consists of screen on, unlock, and a series of app use, and it ends with screen off. (CLICK) Also, there will be notifications from apps which may trigger subsequent usage. (CLICK) We perform fine-grained usage analysis by examining various usage metrics.

9Usage Difference AnalysisOverall usage pattern analysisCategory-specific usage pattern analysis

Self-report AddictionQuestionnaire

Real Usage Data Collection

Exit Interviews(usage patterns)

ContentAnalysisAnalysis of Between-groupUsage Differences

RiskNon-riskQuantitative analysisContent analysisIn our quantitative analysis, we first perform overall usage pattern analysis, and then category-specific usage pattern analysis. 10Usage Difference AnalysisOverall usage pattern analysisCategory-specific usage pattern analysisOverallUsagePatternsAggregated UsageUsage time per daySession frequency per daySession-level UsageSession time (usage time per session)Inter-session time Number of apps used per sessionEntropy of top-k apps usage time/frequency distributions (k = 5, 10, 50)Usage time and frequency of top 1/2 appsDiurnal UsageUsage time, session frequency, and session time: night (0~6), morning (6~12), afternoon (12~18), and evening (18~24)

For overall usage analysis, we examined various metrics such as aggregated usage, session-level usage, and diurnal usage patterns. In the following, I report the key findings. 11Overall Usage DifferencesUsage time & frequencyDaily usage amount: risk 253.0 vs. non-risk 207.4 (p=.011, Cohens d=0.45)

Daily usage freq.: risk 111.5 vs. non-risk: 100.1 (p=.146, Cohens d=0.31)

0120240360480(min)usage amount (per day, per user) 0

100200usage frequency(per day, per user)Risk group spent more time than non-risk group, tending to use devices more frequently & to engage in longer sessions First of all, we examined whether there are any differences in daily usage time and frequency. Here usage frequency means the number of usage sessions per day.

Our data showed that (CLICK) risk group spent more time on their smartphones than the non-risk group. They also tended to use their smartphones more frequently and to engage in longer session (although non-significant) .

While the time difference is not large, psychology literature says that problematic usage is more related to usage functions and purposes than usage amount. 12Overall Usage DifferencesSkewness of app usageUsage of top-k apps skewed (exponential dist)

Notable differences found on usage of top-1/2 apps1st app (risk 97.8 vs. non-risk 69.9 m, p=.003, Cohens d=0.66)2nd app (risk 47.4 vs. non-risk 37.5 m, p=.058, Cohens d=0.43)

Entropy of top-5 apps differs (risk 1.85 vs. non-risk 1.96)

More skewed usage observed in the risk group users In the session-level analysis, we had an interesting finding on skewness of app usage. In general, a persons app usage follows an exponential distribution, meaning that popular apps dominate overall usage.We found notable differences on usage time of top 1/2 apps.For instance, the usage time of top 1 apps differs about 30 minutes.

We can compare the skewness of apps using the entropy measure; and our analysis showed that when we consider top 5 apps, we observe statistically significant differences.

(CLICK) Here, we found that the risk group users show more skewed usage in top-k apps 13Overall Usage DifferencesDiurnal usage differences

Usage TimeDiurnal usage difference exists: i.e., Risk group spent more time during morning/evening RiskNon-RiskWe also found there is a diurnal usage pattern difference. We divide a day into 6 hour blocks; and plot the usage time in each time block.

(CLICK) We found that the risk group spent more time in the morning and evening hours; and overall it appears that risk group users spent consistently more time throughout the day. 14Usage Difference AnalysisOverall usage pattern analysisCategory-specific usage pattern analysisCategory-SpecificUsagePatternsComm.Usage time and app frequency of communication, usage time and freq. of {MIM, SMS, email, call}, inter-MIM timeDiurnal usage: usage time, app frequency, and session time of communication apps: night (0~6), morning (6~12), afternoon (12~18), and evening (18~24)Internal vs. external sessions: number of sessions per day, app sequence length per day, session duration per day/ session, app sequence length per sessionWeb BrowserUsage time and app frequency of web, web content, inter-web timeDiurnal usage: usage time, app frequency, and session time of web-browsing apps: night (0~6), morning (6~12), afternoon (12~18), and evening (18~24)

We then analyzed content category-specific usage patterns where we limited our analysis to communications and web browsing. 15Category-specific Usage DifferenceCategory-specific usage: risk group tended to use longer, but significant difference observed only in web

* Used a simplified app category based on existing app categories from app stores

usage amount (minutes)Com. Web SNS Game Media Launcher Misc. Time

We divide apps into comm, web, sns, game, media, app lunacher; For category-specific analysis, we manually coded the category of 2000 apps used by our participants When considering category-specific usage amount, (CLICK) we found significant usage differences only in web browsing. Usage amount of comm apps show some tendency,but the difference was not significant. 16Comm. Usage Differencescomm. app usage Mobile instant messaging (MIM) is dominatingKakaoTalk is the most popular MIM in Korea MIM usage time (Risk: 75.6 vs. Non-Risk: 65.8, n.s.)MIM usage freq (Risk: 91.2 vs. Non-Risk: 76.9, n.s.)

To further analyze communication usage, we first take a look at the popularity of apps.

(CLICK) Our undergrad participants used mobile instant messaging by far more than any other communication channels. In Korea popular mobile instant messaging is KakaoTalk and LINE. In terms of MIM usage, our dataset did not show significant differences. 17Comm. Usage DifferencesExternally-triggered session usageWill externally-triggered session usage differ? Major external triggers: mobile instant messaging (KakaoTalk), SMS, calls, facebook (mostly comm. category) # notifications per day: Risk: 451.8 vs. non-risk: 378.5 (n.s.)External triggering by app1: notification (including incoming calls)app1app2timeapp2Here, if the subsequent session includes app1, we assume that app1 triggered smartphone usage

One of the key concepts related to behavioral addiction is that usage can be trigged by external cues.In our study, we examined whether external triggers like notifications from instant messaging, SMS, and Facebook influence the usage patterns.

(CLICK) For example, app1 has a notification (CLICK) and if the subsequent session includes app1, (CLICK) then we can simply assume that app1 actually triggered smartphone usage. In our dataset, notifications are mostly from kakaotalk; and its number is huge; in the case of risk group, it is surprisingly more than 400. 18Comm. Usage DifferencesExternally-triggered session usageSignificant differences on MIM-triggered sessions Aggregated app seq. length per day (risk: 88.7 vs. non-risk: 63.9, p=.031, Cohens d=0.50)Aggregated session duration per day (s) (risk: 4978.9 vs. non-risk: 3661.5, p=.030, Cohens d=0.50) MIM acts as external cues for usage, causing overuseWe found significant differences on mobile instant messaging triggered sessions.In particular, we find that risk groups aggregated app sequence length, and aggregated session duration is much longer than that of the non-risk group.

(CLICK) From this we can say that MIM acts as external cues of smartphone usage, which may cause smartphone overuse. 19Web Usage DifferencesSignificant difference in usage amount (risk: 67.14 m vs. non-risk: 41.14 m, p=.012, Cohens d=0.61)Selected each users top-10 visited sites and manually classified sites (e.g., portal, search, forums, news)Significant differences in web portal usage and trending issue search (more frequent visits by risk group) Excessive page visits by several risk group users (checking updates of online communities; e.g., 414, 166, 97 pages per day)

Risk group spent more time on the web consuming various types of online contentWe also examine web usage; and there was a significant difference in usage amount.We manually classified top 10 most visited sites of each user to see whether there are any differences in content consumption patterns.We found that there are significant differences in web portal usage, and trending issue search.Also, we observe that several risk group users showed excessive web page visit patterns for checking updates in online communities.

(CLICK) Overall, our risk group participants spent more time on the web, consuming various types of online content. 20Usage Difference SummarySpending more time with their smartphones Diurnal usage differsMore skewed app usage More usage on MIM-trigged sessions (external cues)Spending more time on online content consumption

Risk group usageTo summarize usage differences, we found that the risk group usersSpent more time with their smartphones,Showed diurnal usage differencesTheir top-k app usage was more skewedShowed more usage on instant-messaging-triggered sessionsAnd overall spending more time for online content consumption

21Themes of Problematic UsageInterview data analysisTo supplement usage analysis results and to gain better understanding about usage behavior related to smartphone overuse

Self-report AddictionQuestionnaire

Real Usage Data Collection

Exit Interviews(usage patterns)(4 risk, 3 non-risk participants)

ContentAnalysisAnalysis of Between-groupUsage Differences

RiskNon-riskQuantitative analysisContent analysisWe then performed interviews to corroborate our findings and to gain better understanding about usage behaviors related to smartphone overuse. We interviewed 4 risk and 3 non-risk participants whose usage hours are more than four hours a day. 22Themes of Problematic UsageInterview data analysisFelt more compelled to check their smartphones

I keep paying attention, because I feel like new messages may have arrived.

When Im dating or hanging out with friends, KakaoTalk messages make me feel nervous.

In the interest of time, we report two themes of problematic usage. One interesting theme is that risk group users felt more compelled to check their smartphones,

(CLICK) One participant said, .(CLICK) Another participant commented about nervous feeling related to instant messaging, by saying ..Our results extend -s finding about habitual checking behavior in that frequent interruptions due to instant messages make people more compulsive for checking.

23Themes of Problematic UsageInterview data analysisLess conscious/structured smartphone usage behavior (showing lack of self-regulation)

Its not like I plan to use my smartphone, but I just turn on my smartphone unconsciously. I once used my smartphone to wake myself up in the morning. I got up at 9AM, but it turned out it was already 11AM.

I dont have any thoughts when using my smartphone. ... At that moment, [Im] without any sense of time.Another theme is that risk group users are less conscious, and structured in terms of their smartphone usage, showing lack of self-regulation.

(CLICK) One risk user said, . (CLICK) Another participant EVEN commented about losing sense of time, by saying

24Themes of Problematic UsageInterview data analysisLess conscious/structured smartphone usage behavior (showing lack of self-regulation)

I use my smartphone for about 20 minutes before going to bed. I take care of messages piled up in KakaoTalk, and check Facebook, and webtoon updates at 11:30 PM. After checking that, Im done! (Non-risk group user)

In contrast, our non-risk participants are quite conscious about their usage patterns,

(CLICK) for instance, one participant said, ..

This pattern is very different from that of our risk group participants who were often not able to elaborate their usage behavior. 25Automatic Identification Apply machine learning techniques to test the feasibility of the risk-group classificationUsage features: General set contained the general usage features such as the usage time/frequency, top-k entropies, and sequence lengthTemporal set included the diurnal usage features for aggregated usage and category-specific usageCategory set included the category-level usage features (no temporal aspects)External set included all of the usage features for the external sessionsAutomatic Identification Category-specific usage patterns enable accurate identification of risk-group users

Weka v3.6: Decision Tree (DT), Naive Bayes (NB), and Support Vector Machine (SVM)Feature selection: Information gain algorithmConclusionIdentified usage patterns related to problematic usageUsage amount, skewness of top-k apps, diurnal usage, MIM-triggered usage, web usageUncovered common themes of problematic usage Compulsion of checking updates Limited self-regulation (less conscious/structured usage)Demonstrated the feasibility of automatically identifying risk-group users via machine learningCategory-specific usage enables accurate classification Ongoing Work:Doing a comparative analysis of different datasets Designing intervention s/w for smartphone overuseTo summarize, we identified unique usage patterns related to problematic usage; and uncovered common themes of problematic usage.

Some of the ongoing works include 1) Doing a comparative analysis of different datasets; --- we were able to observe similar patterns from another dataset. 2) Designing intervention s/w for smartphone overuse.

With that I thank the audience, and receive some questions. 28

Hooked on Smartphones: An Exploratory Study on Smartphone Overuse among College StudentsUichin Lee, Joonwon Lee, Minsam Ko, Subin Yang, Gahgene Gweon (KAIST, Knowledge Service Engineering) Changhun Lee, Yuhwan Kim, Junehwa Song (KAIST, Computer Science) Koji Yatani (Microsoft Research Asia)Kyong Mee Chung (Yeonsei University, Dept. of Psychology)