NAIST-IS-MT1351001 2015 3 12
( )
LINE Twitter Web, NAIST-IS-MT1351001, 2015 3 12. i
LINE 9 5 1 1 3 50 % 20 % 5 72 %,,,, ii
A Method for Estimating Appropriate Interruption Timings using Recipient s Context Yuma AKAIKE Abstract Recently, the spread of smartphones and mobile network are promoting a real time communication such as LINE and Twitter. Not only these applications but also various applications such as a whether information or social network services have a function of push notification. As a result, we receive many notifications from various applications. These notifications are useful for recognizing the up-todate information, but frequent interruption can cause a decrease in concentration as well as stress. The goal of this thesis is to estimate an appropriate timing of interruption for notifying something to the recipient. In order to achieve it, We try to recognize a human activity via a smartphone. After showing its feasibility, we try to create a model for estimating a future response time by analyzing the recipient s contexts. In addition, we try to create an another model for estimating a recipient s busyness. At first, we collect the human activity log and the text history of smart-phone applications. Then, we investigate whether those data can be used to infer recipient s context. As a result, we found the recipient s activity inference can be estimated to some extent from those data. Master s Thesis, Department of Information Science, Graduate School of Information Science, Nara Institute of Science and Technology, NAIST-IS-MT1351001, March 12, 2015. iii
Based on the result of investigation, we conducted the experiment in real environment by using LINE. Various tasks are sent to 9 subjects in various timings. All the subjects did not know anything about the experiment in advance. We measured the response time for each task. Finally, I ask every subject about the actual context when they received a message. Their contexts were recorded by a life logging application Moves. By using these information, We created two models, response time estimation model and busyness estimation model, by using machine learning. As a result, we confirmed that our proposed models can estimate a response time over 50 % accuracy in the case we classify the response time into 3 level ( within 5 minutes, within 60 minutes and over 60 minutes ) and our proposed models can estimate a recipient s busyness level with over 70 % accuracy also in the case I classify it into 5 levels. Keywords: Interruption, Life-log, Data-mining, User-context, Machine learning iv
1. 1 2. 5 3. 10 3.1................................ 10 3.1.1.................... 10 3.1.2.................... 12 3.2................................ 12 3.3.................................. 16 3.4................................. 18 4. 20 4.1................... 20 4.1.1... 20 4.1.2.......... 21 4.2........... 22 4.2.1............... 23 4.2.2.................. 23 4.2.3................... 24 4.2.4.... 25 4.2.5 27 4.3................... 29 4.3.1..................... 29 4.3.2..................... 31 4.4............................. 32 4.4.1....................... 33 4.5................. 33 4.6.................................. 35 v
5. 37 5.1............................ 37 5.2...................... 38 5.2.1..................... 38 5.2.2....................... 39 41 42 vi
1...................... 7 2 1 A. 15 3 1 B. 16 4 5 A....................... 17 5 5 B....................... 18 6................. 24 7...................... 25 8....................... 26 9 2................................. 28 10 2.............................. 28 11............ 32 1................. 22 2........................ 30 3........................ 31 4................... 35 vii
1. LINE 1 Twitter 2 PC [1] Garrett Danziger [2]. Cutrell [3] [4]. 1 www.line.naver.jp 2 www.twitter.com 1
[5][6] 1. 2. 2 2 1 GPS 1 Web 2
2 Fitibit Zip 3 1 5 Line Twitter LINE -0.79 Twitter -0.94 LINE LINE 9 3 3 www.fitbit.com 3
Moves 259 SVM RandomForest J48 [7] 46.3 F 5 1 1 3 71.8 53.3 3 20 4
2.. Garrett Danziger argue [2] Cutrell [3] Fischer [8] PC Avrahami Hudson PC [9] 5
Horwitz [10] PC Ter Hofte [11]. [12] [13] model-based produres [14] Web [15] 6
Pielot [16] Veljko Happy Sad Bored [17] 1 Veljko 1: 7
Gmail Mail SNS SNS 140 Twitter Facebook messenger 4 4 https://www.facebook.com/ 8
[18] 9
3. 3.1 2 30 Fitbit Zip SNS LINE Twitter Fitbit Zip 1.4 1.1 8 4 6 3.1.1 1 LINE Twitter SNS Facebook Gmail Facebook Gmail 1 10
Twitter API 5 Python 5 LINE Twitter API.txt.txt Python SNS 1. LINE 2. Twitter 5 https://dev.twitter.com/ 11
3.1.2 Fitbit Fitbit Zip Fitbit Zip 5 Fitbit Zip bluetooth Web Fitbit API Fitbit API 5 LINE Twitter Fitbit Zip Python matplotlib 5 3.2 2 1 30 LINE Fitbit Zip 2 3 A 12
B Fitbit LINE Twitter 1 10 22 1 Log count = (raw data of Log count) (Maximum raw data of Log count) 2 A 3 B LINE Twitter. 1 13 (1)
1 0 0 LINE -0.79 Twitter -0.94 0 0.1 Twitter LINE Foursquare 6 1 2 3 5 0 1 100 101 200 201 300 301 400 5 4 A 5 B x 5 x y y 6 ja.foursquare.com/ 14
10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 2: 1 A LINE Twitter 4 5 200 15
10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 3: 1 B 3.3 16
4: 5 A Gmail Evernote 7 Facebook Twitter 7 www.evernote.com 17
5: 5 B 3.4 18
1 LINE Twitter [19] 19
4. 3 SNS 4.1 9 23 24 7 2 2 4.1.1 3 1 LINE 20
2 24 25 1 1 3 1 9:00 11:59 12:00 14:59 15:00 17:59 1 4.1.2 2 Moves 8 Moves API Moves 8 https://www.moves-app.com/ 21
Moves Moves 5 Moves Moves 1: 1 Yes or No??? 2??? 3?? 4.2 22
4.2.1 6 1 3 2 3 2 4.2.2 7 1 23
1:26 1:12 0:57 0:43 0:28 0:14 0:00 レベル 1 レベル 2 レベル 3 6: 4.2.3 8 2 4 5 3 1 2 1 2 24
1:55 1:40 1:26 1:12 0:57 0:43 0:28 0:14 0:00 朝昼夕方 7: 1 2 4 5 4.2.4 3 25
2:24 2:09 1:55 1:40 1:26 1:12 0:57 0:43 0:28 0:14 0:00 とても忙しい 忙しい どちらでもな い 暇 とても暇 8:???? 5 2?? 5 1 4 2???? 2 4 2 1???? 2?? 1 3 2 26
?? 1 3 4.2.5???? 9 10 2 9 27
1:40 1:26 1:12 0:57 0:43 0:28 朝 昼 夕方 0:14 0:00 忙しい 暇 9: 2 5 4 3 2 朝 昼 夕方 1 忙しい 暇 10: 2 28
4.3 F 4.3.1 6 SVM RandomForest J48 1 3 AM 9:00 AM 11:59 PM 12:00 PM14:59 PM 15:00 PM 17:59 PM 18:00 PM 21:00 4 2 3 Moves 2 29
2: 30
259 1 173 3: 4.3.2 5 5 1 2 8 5 31
11: 11 4 5 SVM 36.9% J48 37.4% RandomForest 37.0% 3 5 1 SVM 36.9% J48 46.3% RandomForest 43.0% 4.4 32
4.4.1 1 3 AM 9:00 AM 11:59 PM 12:00 PM14:59 PM 15:00 PM 17:59 PM 18:00 PM 21:00 4 2 3 Moves 2 5 SVM RandomForest J48 SVM 42.2% RandomForest 71.8% J48 63.3%. 4.5 5 71.8% 1 3 33
AM 9:00 AM 11:59 PM 12:00 PM14:59 PM 15:00 PM 17:59 PM 18:00 PM 21:00 4 2 3 Moves 2 5 SVM 53.3% RandomForest 49.0% J48 50.0% 13% 3 33% 20 34
4: 5 3 2 5 3 J48 precision 35.7% 47.3% 41.6% 63.8% 50.7% recall 41.3% 47.1% 64.5% 65.1% 50.5% F-measure 37.4% 46.3% 50.6% 63.3% 50.0% SVM precision 31.9% 32.1% 41.6% 65.7% 53.9% recall 43.6% 43.6 % 64.5% 54.1% 53.4% F-measure 36.9% 36.9 % 50.6% 42.2% 53.3% Random precision 36.2% 43.0 % 64.5% 71.4% 48.9% forest recall 38.4% 43.0% 65.7% 72.7% 49.0% F-measure 37.0% 43.0% 64.8% 71.8% 49.0% 35.5% 35.5% 35.5% 22.1% 35.5% 4.6 LINE Moves 1 35
173 6 3 5 70% 5 1 1 20 51 Moves 36
5. 5.1 Fitbit Zip 5 LINE Twitter 2 30 LINE Moves Yes or No 37
3 1 173 6 3 5 1 1 20 % 53.3 % 5 70% 5.2 5.2.1 53.3 3 38
1 Twitter LINE API, Gmail 5.2.2 2 9 Moves Moves Moves 39
Moves 40
41
[1] Leslie A Perlow. The time famine: Toward a sociology of work time. Administrative science quarterly, Vol. 44, No. 1, pp. 57 81, 1999. [2] R Kelly Garrett and James N Danziger. Im= interruption management? instant messaging and disruption in the workplace. Journal of Computer- Mediated Communication, Vol. 13, No. 1, pp. 23 42, 2007. [3] Edward Cutrell, Mary Czerwinski, and Eric Horvitz. Notification, disruption, and memory: Effects of messaging interruptions on memory and performance. 2001. [4] Brian P Bailey and Joseph A Konstan. On the need for attention-aware systems: Measuring effects of interruption on task performance, error rate, and affective state. Computers in Human Behavior, Vol. 22, No. 4, pp. 685 708, 2006. [5] Hiroshi Takeuchi, Naoki Kodama, Takeshi Hashiguchi, and Doubun Hayashi. Automated healthcare data mining based on a personal dynamic healthcare system. In Engineering in Medicine and Biology Society, 2006. EMBS 06. 28th Annual International Conference of the IEEE, pp. 3604 3607. IEEE, 2006. [6] Nobuo Kawaguchi, Nobuhiro Ogawa, Yohei Iwasaki, Katsuhiko Kaji, Tsutomu Terada, Kazuya Murao, Sozo Inoue, Yoshihiro Kawahara, Yasuyuki Sumi, and Nobuhiko Nishio. Hasc challenge: gathering large scale human activity corpus for the real-world activity understandings. In Proceedings of the 2nd Augmented Human International Conference, p. 27. ACM, 2011. [7] Michael J Berry and Gordon Linoff. Data mining techniques: for marketing, sales, and customer support. John Wiley & Sons, Inc., 1997. 42
[8] Joel E Fischer, Chris Greenhalgh, and Steve Benford. Investigating episodes of mobile phone activity as indicators of opportune moments to deliver notifications. In Proceedings of the 13th International Conference on Human Computer Interaction with Mobile Devices and Services, pp. 181 190. ACM, 2011. [9] Daniel Avrahami and Scott E Hudson. Responsiveness in instant messaging: predictive models supporting inter-personal communication. In Proceedings of the SIGCHI conference on Human Factors in computing systems, pp. 731 740. ACM, 2006. [10] Eric Horvitz and Johnson Apacible. Learning and reasoning about interruption. In Proceedings of the 5th international conference on Multimodal interfaces, pp. 20 27. ACM, 2003. [11] G Henri Ter Hofte. Xensible interruptions from your mobile phone. In Proceedings of the 9th international conference on Human computer interaction with mobile devices and services, pp. 178 181. ACM, 2007. [12] A Rogier T Donders, Geert JMG van der Heijden, Theo Stijnen, and Karel GM Moons. Review: a gentle introduction to imputation of missing values. Journal of clinical epidemiology, Vol. 59, No. 10, pp. 1087 1091, 2006. [13] Patrick Royston. Multiple imputation of missing values. Stata Journal, Vol. 4, pp. 227 241, 2004. [14] Jonathan Marchini, Bryan Howie, Simon Myers, Gil McVean, and Peter Donnelly. A new multipoint method for genome-wide association studies by imputation of genotypes. Nature genetics, Vol. 39, No. 7, pp. 906 913, 2007. [15],,.. 21, 2013. 43
[16] Ruth Rettie, Ursula Grandcolas, and Bethan Deakins. Text message advertising: Response rates and branding effects. Journal of Targeting, Measurement and Analysis for Marketing, Vol. 13, No. 4, pp. 304 312, 2005. [17] Veljko Pejovic and Mirco Musolesi. Interruptme: designing intelligent prompting mechanisms for pervasive applications. In Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing, pp. 897 908. ACM, 2014. [18] Jon C Hammer and Tingxin Yan. Exploiting usage statistics for energyefficient logical status inference on mobile phones. In Proceedings of the 2014 ACM International Symposium on Wearable Computers, pp. 35 42. ACM, 2014. [19],.. In DEIM Forum, 2009. 44