On Cross-Domain Data Access for Cyber-Physical-Social S ystems - PowerPoint PPT Presentation

On Cross-Domain Data Access for Cyber-Physical-Social S ystems Koji Zettsu zettsu@nict.go.jp Universal Communication Research Institute National Institute of Information and Communications Technology (NICT) International S ymposium on Social M ultimedia and M ultimedia Computing CASIA, China August 15-16, 2013

Cyber-Physical-Social Information Ecosystem Environmental Smarter society & Disaster response application science Life innovation Analysis Cyber Space Service Gathering Sensing Actionable feedback Ubiquitous Network 2 SNS Physical Space Science databases Sensor networks Web 2 Natural environments Disasters Traffics Phenomena Social activities Life events 2013/ 8/ 15 (C) NICT 2

Collective Awareness of Cyber-Physical Social Event (E.g.) PM 2.5 air pollution PM 2.5 distributions Environment Sensing PM 2.5 treatments Profile Gathering Resource Availability Locations Sensitivities Analysis/ Prediction Individual Interaction Individual Response Disaster risk analysis • Climate influences • Disaster response communities • • Health influences Social healthcare Recommendation Action/Advice Social Response Courtesy: Event information management system 2013/ 8/ 15 (C) NICT 3 (UC Irvine, NICT, 2013)

Cyber-Physical Social System: A Vision Recognize evolving complex situations at particular location over period of time, then provide actionable feedback to device, people, and society M odel and manipulate cyber-physical-social events based on ‘correlations’ between individually- disseminated, heterogeneous sensor data 2013/ 8/ 15 (C) NICT 4

Analyzing Spatial-Temporal-Thematic Correlations High PM 2.5 (> 35 μg / m 3 ) High humidity (> 85%) Twitter (“face mask”) “I need to ware face mask, because PM 2.5 increases extraordinarily” “Wearing face mask makes my face smaller in such a humid day” 2013/ 8/ 15 (C) NICT 5

STICKER Spatio-Temporal Information Clustering and Knowledge ExtRaction Visual correlation mining of heterogeneous sensor data in spatial-temporal-thematic (STT) space •Discover correlating dataset Infoglut! •Narrow STT conditions Visual analysis of STT correlations Visual presentation of intersection, surrounding, synchronization, movements and continuity complement, etc. Aggregation Data Filtering Filtering Filtering M anipulation 集約 Sensor STT Cell STT Cell Data 変換 data Composite M odel 変換 Visualization Trajectory T ag cloud Plotting Iso-sphere Composite 2013/ 8/ 15 (C) NICT 6

NICT K-L Grid: Cyber-Physical Social Sensing Platform E n v i r o n m e n t P e o p l e H e a l t h c a r e User Hokuriku Node Okayama Data aggregation Data warehouse J oin Participatory Grid Network Fukuoka User Tokyo Node F a c i l i t i e s S c i e n c e Kyoto New Generation Network/JGN-X testbed (C) NICT 2013/ 8/ 15 7

Sensing Platform Issue Agile gathering, processing and archiving of heterogeneous sensor data Application developer Collective sensing application • Sudden deluge of data from On-demand creation of anonymous sources • Ad-hoc sensing and trial- application-specific error analysis virtual sensor networks • Quick response to ongoing situation changes Physical networks • Suffer from unforeseen traffics Network • Hard to reconfigure existing administrator sensor networks 2013/ 8/ 15 (C) NICT 8

Cyber-Physical-Social Sensor Service (CPSenS) • Sensor Service Collaboration Sensor Service Collaboration Overlay – Sensor virtualization : encapsulates data sources as sensor services – Vertical sensor integration : combines heterogeneous sensor services on demand – Horizontal sensor integration : complements 1. Declarative Service Networking missing data with multiple sensors 3. DSN/ NCPS SCN Translator • Service Controlled Networking (SCN) 1. Declarative Service Networking (DSN): 2. Network Control Protocol Stack defines application-specific sensor service NCPS for NCPS for collaboration by declarative rule language OpenFlow IEEE 1888 2. Network Control Protocol Stack (NCPS): IEEE 1888 OpenFlow Invoke programmable network commands Command/API Command/API for dynamic configuration; service node discovery, path setting, status monitoring DSN/ NCPS Translator: Generate NCPS 3. commands by interpreting DSN OpenFlow NW IEEEE 1888 NW descriptions with multiple-overlay coordination Programmable Networks 2013/ 8/ 15 (C) NICT 9

CPSenS Example DSN description Overlay GeoSocialApp // 1. Service registration R1 REGIST(GeWE) <~ IDENT(GeoSocialWeb, GeWE, “192.168.94.62”) // 2. Service search F1 FIND(RaQU) <~ REQUEST(GeWE, RaQU) F2 FIND(TwQU) <~ REQUEST(GeWE, TwQU) // 3. Path creation and message exchange S1 SEND(GeWE, RaQU, “50 <= Rain” & “1000 < SampleRate”) <~ FIND(RaQU) S2 SEND(GeWE, TwQU, “disaster” & “4000 < Record”) <~ FIND(TwQU) OpenFlow paths by NCPS Sensor service collaboration 2013/ 8/ 15 (C) NICT 10

Seamless Processing form Sensor Data to Complex Event “aggregation” operator + “ filter” Air pollution event operator Event model (#e1 + #e2 > b) “ filter” topic operator (#e3 > a) func space time tion S Abnormally increasing of object PM 2.5 (e1) o o “ wearing mask” topic e e e (e3) Abnormally decreasing of wind speed o o o Longitude, (e2) latitude, time d d d O: operators TOPIC ABNORM AL E: event detection detection S: situation D: sensors’ data PM 2.5 sensor Courtesy: Event information management system (UC Twitter sensor Wind speed sensor Irvine, NICT, 2013) 2013/ 8/ 15 (C) NICT 11

Event Information M anagement System Trigger Complex event processing Event Shop (UCI) Analyzing complex situations by spatio-temporal event Reuse stream processing Event Warehousing EvWH (NICT) Creating atomic events from C-P-S Event DB individually-disseminated , Event heterogeneous sensor data Link detection & correlation (both streams and archives) Sensor service Sensor data archives Courtesy: Event information management system (UC Irvine, NICT, 2013) 2013/ 8/ 15 (C) NICT 12

Event Warehouse (EvWH) Correlation Database (Value-based Storage approach) Weather sensor data SNS sensor data (weather topic) Timestamp Station Temperature precipitation Datetime Geotag #Tag Text 2012-03-01 11:02:35 AP-Kyoto 2.1 C 3.5mm 2012-03-01 11:00:02 35.011636, 135.768029 #Weather Wondering if it snows 2012-03-01 11:03:05 AP-Tokyo 11.2C 0.1 mm 2012-03-01 11:02:00 35.681382, 139.766084 # Weather Spring has come 2012-03-01 11:03:35 AP-Osaka 1.0C 2.1mm 2012-03-01 11:05:45 34.701909, 135.494977 #Weather Not good for laundry Data values Table structure Data values Table structure tag:value record_tag:[tag, tag,… ], tag:value record_tag:[tag, tag,… ], 1:2012-03-01 11:02:35, 1001:[1, 4, 6, 9], 21:2012-03-01 11:00:02, 2001:[21, 24, 27, 28], 2:2012-03-01 11:03:05, 1002: [2, 5, 7, 10], 22:2012-03-01 11:02:00, 2002: [22, 25, 27, 29], 3:2012-03-01 11:03:35, 1003:[3, 4, 8, 11] 23:2012-03-01 11:05:45, 2003:[23, 26, 27, 30] 4:AP-Kyoto, 24:35.011636, 135.768029, 5:AP-Tokyo, 25:35.681382, 139.766084 , 6:2.1C, 26:34.701909, 135.494977, Ontological correlation 7:11.2C, 27:#Weather, (metrology) 8:1.0C, 28: Wondering if it snows, 9:3.5mm, 29: Spring has come, 10:0.1mm, 30: Not good for laundry 11:2.1mm Ontological correlation (Atmosphere) Spatiotemporal correlation JOIN over Correlation Correlation corr_tag: (tag, tag, ..)/ corr_coeff … corr_label Table structure 101: (1, 2, 3, 21, 22, 23)/ 0.95 … [2012-03-01 noon] record_tag:[tag/ certanty, tag/ certainty,… ],) 102: (4, 24, 26)/ 0.92 … Kansai area 5001:[101/ 0.95, 102/ 0.92, 104/ 0.85, 106/ 0.6,], 103: (5, 25)/ 0.80 … Tokyo area 5002: [101/ 0.95, 103/ 0.80, 105/ 0.9, 29:/ 1.0] 104: (6, 8, 9, 11)/ 0.85 … snow 105: (7, 10)/ 0.9 … sunny Period Area Weather Atmosphere 106: (28, 30)/ 0.6 … cheerless day 101/ 0.95: 102/ 0.92: 104/ 0.85: 106/ 0.6: [2012-03-01 noon] Kyoto area snow cheerless day Cyber-Physical-Social Event 101/ 0.95: 103/ 0.80: 105/ 0.9: 29: [2012-03-01 noon] Tokyo area sunny Spring has come

Correlation Search (Cross-DB Search) Spatiotemporal Ontological Citational Correlation Correlation Correlation Time Long. Lat. Complex Correlation Analysis Correlation Optimization Graph Event M etadata Search 2013/ 2/ 28 (C) NICT (C) NICT 2013/ 8/ 15 14

Correlation Clustering by Evolutional Computing • Optimize correlation graph cluster by evolutional operations: merge, split, expansion, crossover, and mutation Generation 0 Generation 1 Generation 3 Generation 4 Result 1 mutation mutation merge crossover Result 2 split expansion mutation Result 3 Evolutionary tree of correlation graph • The more generation grows, the more correlation graphs become strongly connected (thick edges = strong correlations) (C) NICT 2013/ 8/ 15 15

Example of Cross-DB Search Query ： “ice sheet” in Northern Hemisphere Select spatially- and ontologically- correlating data Examples of added dataset 2013/ 8/ 15 (C) NICT 16