Group Detection Based on User-to-User Distance in Everyday life for Office Lunch Group Recommendation

Transcript

1. Group Detection Based on User-to-User Distance in Everyday life 
 for Office Lunch Group Recommendation NARA Institute of Science and Technology ˕Ryota Koshiba, Yuko Hirabe, Manato Fujimoto, 
 Hirohiko Suwa, Yutaka Arakawa, Keiichi Yasumoto COLLABES 29-3 2017
2. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 2
3. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 3
4. Background In big organization (big company) • people are segmented into many different sections • project members are vertically organized 4 Lack of communication between sections is serious → More active inter-section communication is required
5. Events to promote communication • Companies hold various events to promote communication among people in different sections 5 Lunch meeting Drinking party Sports Festival These hold outside office hours →Possible to cause dissatisfaction of workers More and more companies held an event called “shuffle lunch” →We focus on Lunch meeting
6. Grouping policies for lunch 6 • almost no conversation • stress by too much caring • make no connection between new people Grouping new people only may result in: Grouping good friends only will: hybrid grouping that includes new people and common friends is necessary Knowing actual human relationship is needed
7. Definition of human relationships 7 • Boss⁶Manager • Boyfriend⁶Girlfriend • ColleaguesɺPartnersɺ etc. Physical relationship Social relationship Person-to-person distance Actual human relationship correlates with: the time duration while persons are in their proximity To know human relationships, it is important to grasp person-to-person distance in everyday life
8. Objective of this study 8 Extract human relationship from frequency/ time of people to act together Create a new group based on extracted human relations Step. 1ɹGroup detection Step. 2ɹGroup creation • Recommending a group considering actual human relationships
9. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 9
10. Related Work 10 Existing studies on group detection • GPS is used →Unavailable indoors →power consumption is high →need to upload position data to server (privacy) • Wireless LAN is used →cannot be used in the area outside WiFi AP coverage • Proprietary device is used →difficult to deploy in actual companies GPS Wireless LAN BLE IR Proprietary device Power consumption [1] ✔ ✔ - - Nothing High [2] - - ✔ ✔ Yes Low Proposed method - - ✔ - Nothing Low <>4
    3JKJVSFLIB
    FU
    BM
    (SV.PO
    'BTU
    BOE
    "DDVSBUF
    (SDIBOSPVQ
    .POJUPSJOH
    GPS)FUFSPHFOFPVT
    6SCBO
    4QBDFT
    
    SE
    FE
    
    
    1SPD
    PG
    "$.
    4FO4ZT
    
    
    <>%
    .
    5
    5SJOI
    FU
    BM
    (SPVQ6T
    4NBSUQIPOF
    1SPYJNJUZ
    %BUB
    BOE
    )VNBO
    *OUFSBDUJPO
    5ZQF
    .JOJOH
    
    SE
    FE
    1SPD
    PG
    
    UI
    "OOVBM
    *OUFSOBUJPOBM
    4ZNQPTJVN
    PO
    8FBSBCMF
    $PNQVUFSTQQ
    
11. Requirements for group detection system 11 1. Diffusiveness: Instead of using proprietary devices, only smartphones currently on sale are used 2. Energy efficiency: Energy consumption of the system must be small 3. Privacy: Personal information infringing privacy (like positions) is not sent or received Our approach • BLE (power saving, privacy preservation) • ordinary Android terminal equipped 
 with a BLE peripheral function
12. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 12
13. Proposed system 13 To know person proximities, we measure user-to-user distance in everyday life by using BLE advertisement packets System flow: 1. Send and receive BLE advertisement packets between users 2. Measure distance between users based on BLE RSSI 3. Based on the distance, we judge whether together or not Send and receive 
 BLE signal User-to-user distance
14. System overview 14 To detect groups from user proximities, our system has two components: • Smartphone application for measuring user proximities • Cloud system for detecting groups based on user proximities 3. Save and analysis all collected data 1. Send and receive BLE packets to each other 2. Upload data to cloud 4. Detect groups based on analysis results Group A Group B
15. User Interface of proposed application 15 Specification of Application: • Minimum interval of signal reception: 5 seconds • Always working in the background ᶃTap Button ᶄStart send & scan packet Top Screen Scanning Screen ᶅwhen packet received, display and save four information: • UUID of BLE from both side • RSSI of BLE signal • Received time
16. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 16
17. Evaluation Purpose: 
 To confirm relevant between RSSI and user-to-user distance 17 Purpose: To Confirm effect on RSSI from factors other than user- to-user distance
 ʲNumber of People, Environment, Terminalʳ 1. Basic performance evaluation of BLE signal propagation ̎. Effect on RSSI from different situation To show the effectiveness of proposed system, we carry out the following system two evaluation:
18. Basic performance evaluation of BLE signal propagation • To confirm relevant between RSSI and user-to-user distance, measure while changing distance between terminals 18 • Distance:
 50cm, 100cm, 200cm, 300cm, 500cm, 700cm, 
 1000cm, 1500cm, 2000cm Evaluation flow: 1. Collecting data for 5 minutes at each distance 2. Compare based on average of each data distance [cm]
19. RSSI [dBm] -100 -90 -80 -70 -60 -50 -40 log [cm] 10 100 1000 10000 Relevant between RSSI and user-to-user distance 19 • RSSI gradually decreased as distance increased → Confirmed that can measure user-to-user distance by using RSSI of BLE mounted on smartphone
20. Effect on RSSI from different situation To Confirm effect on RSSI from factors other than user-to- user distance, investigate effect in multiple situations
 Experimental condition: • Participants: 8 students • All participants used smartphone of same model 
 (proposed application installed) Instruction to participants: • Participants acts freely with smartphone • As correct data, record 
 "Time", 
 "Which participants went with you",
 "Where you went" 20
21. Effect on RSSI from different situation • 2 participants, Place: Shop, Purpose: Shopping • 5 participants, Place: Restaurant, Purpose: Lunch
 Confirm effect on RSSI by number of people and terminal 21 • 2 people, Place: classroom 
 (assuming a meeting in the company) 1. Movement by car mainly 2.During class In this experiments, confirm effect by environment
22. Experiment environment by car mainly 22      ` ` ` ళ" ళ# େֶߏ಺  ` றं৔ εϚʔτϋ΢ε ݚڀࣨ ίϯϏχ ໿LN ໿LN ໿ N ໿N University Shop Restaurant Smart House Lab. CVS Parking
23. Movement by car mainly (2 participants) RSSI received by terminal A 23 RSSI [dBm] -90 -80 -70 -60 -50 -40 -30 -20 Time [HH:mm] 13:40 14:01 14:23 14:44 15:06 15:27 15:49 16:10 16:32 16:53 17:15 terminal B Walking Walking Smart house Car Shopping Car work in Lab. • RSSI is larger than other actions • Variations in values is small (standard deviation: 1.29) → depend on seat and closed room space
24. RSSI [dBm] -80 -76 -72 -68 -64 -60 -56 -52 -48 -44 -40 Time [HH:mm] 13:40 14:01 14:23 14:44 15:06 15:27 15:49 16:10 16:32 16:53 17:15 RSSI [dBm] -80 -76 -72 -68 -64 -60 -56 -52 -48 -44 -40 Time [HH:mm] 13:40 14:01 14:23 14:44 15:06 15:27 15:49 16:10 16:32 16:53 17:15 Movement by car mainly (2 persons) 24 • Almost no difference in RSSI
 (correlation coefficient: 0.98) • Dependence between terminals is small →All terminals are unnecessary to receive BLE Compare RSSI received by each other
25. Situation: • 5 participants go to restaurant for lunch Action Flow: • 4 participants move by car, 1 participant moves on scooter • After lunch, 3 participants go shopping to convenience store Movement by car mainly (5 participants) 25 School Restaurant
26. Movement by using car mainly (5 people) 26 RSSI [dBm] -100 -90 -80 -70 -60 -50 -40 -30 Time [HH:mm] 11:50 12:00 12:11 12:21 12:32 12:43 12:53 13:04 13:15 terminal B terminal C terminal D terminal E Car Car Eating Only terminal E another move RSSI received by terminal A Walking Walking CVS Go to CVS 
 with terminal B,D • Correlation coefficient between A and B: 0.93 (for two people: 0.98) • Standard deviation of car movement: 2.30 
 (2 people: 1.29) • → Possible to detect groups even if terminal increased
27. RSSI [dBm] -100 -95 -90 -85 -80 -75 -70 -65 -60 Time [HH:mm] 9:00 10:00 11:00 12:00 13:00 14:00 15:00 sitting next B sitting little far away act separately from B Join in 
 laboratory During class (2 participants) 27 First lesson Second lesson Lunch time and third lesson • Confirmed RSSI data can be separated by each situation
 (whether together or not)
28. OUTLINE • Background • Related Work • Proposed System • Evaluation • Conclusion 28
29. Conclusion Group detection based on human relationship: • use BLE RSSI to know user-to-user distance • measured RSSI of users who acted together in many different situations Findings: • Group detection is possible by BLE • RSSI is different depending on situations 
 → possibility of estimating behavior 
 → better human relationship estimation 29 Future Work: • Determine conditions for group detection • Evaluation of group detection accuracy • Constructing a group recommend, system based on human relations
30. Thank you for your kind attention