© Copyright 2010 Digital Enterprise Research Institute. All rights reserved.
Digital Enterprise Research Institute www.deri.ie
Towards Unified and Native Enrichment in
Event Processing Systems Souleiman Hasan, Sean O’Riain, Edward Curry
Digital Enterprise Research Institute, National University of Ireland, Galway, Ireland
In proceedings of The he 7th ACM International Conference on Distributed Event-Based Systems June 29 - July 3, 2013, Arlington, Texas, USA
[email protected] http://www.StefanDecker.org/
Digital Enterprise Research Institute www.deri.ie
Talk Overview
n Introduction ¨ IoT, Cyber-Physical Systems
¨ Event Incompleteness
n Current Approaches n Proposed Approach
¨ Challenges for Enrichment
¨ Proposed Model
¨ Implications
n Linked Data Instantiation ¨ Evaluation
n Summary and Future Directions
2/30
Digital Enterprise Research Institute www.deri.ie
Big Data & IoT
Digital Enterprise Research Institute www.deri.ie
Cyber-Physical Systems
Smart City Smart Grid
Smart Building Smart Enterprise
Digital Enterprise Research Institute www.deri.ie
Event Processing Systems
n Three dimensions of decoupling
n Removal of explicit dependencies between event producers and consumers ð Scalable deployment
n Information exchange only by Events
n (Eugster et al., 2003)
Space
Time
Synch Producer Consumer
5/30
Digital Enterprise Research Institute www.deri.ie
Problem – Event Incompleteness
n Event producers and consumers are decoupled ð Event producers may have very little knowledge about
consumers information needs
Environmental Sensors (Event Observers/Producers)
{(type= "energy consumption") and (floor= “first floor") and (consumption="high")}
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”) Event Processing Engine
Business User
6/30
Digital Enterprise Research Institute www.deri.ie
Dimensions of Incompleteness
n Event Format: lacks syntactical structure ¨ E.g. plain text against conjunctive subscription
n Event Semantics: events lack a reference scheme ¨ E.g. schema-less tuples
n Lack of Background Knowledge ¨ E.g. complementary information exists in external DB
n Incompleteness Addressable by Transformation ¨ E.g. transforming amounts of multiple measurement units
n Temporal Segmentation ¨ E.g. Complementary information exists in past or future
events
7/30
Digital Enterprise Research Institute www.deri.ie
Current Approaches
8/30
Digital Enterprise Research Institute www.deri.ie
Event Enrichment
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(room, “202e”)
(floor, “second floor”)
9/30
DEVICE ROOM FLOOR Color
heater x 202e second floor white
heater y 313 third floor blue
Meta Data
IoT Heater Event
Enriched Event
Digital Enterprise Research Institute www.deri.ie
n Dedicated agents to complement events
Agent-based Event Enrichment
Producer
Producer
Rule1
Event Processing Agent
Enricher
Enricher
Consumer
10/30
Digital Enterprise Research Institute www.deri.ie
n Pros ¨ Events complete with respect to consumer’s need
¨ Low false positives/negatives rate
n Cons ¨ Ad-hoc and external to event processing engines
¨ Difficult to develop and maintain enrichment logic
¨ Difficult to optimise enrichment process
Agent-based Event Enrichment
11/30
Digital Enterprise Research Institute www.deri.ie
Proposed Approach
12/30
Digital Enterprise Research Institute www.deri.ie
Proposed Approach
n We need ¨ Event enrichment to be integrated into the event
processing paradigm as a core task of event processing engines
n Proposal ¨ Unified declarative language for event processing
and enrichment
¨ Enrichment element as a declarative specification for engine to enrich events with complementary information items
13/30
Digital Enterprise Research Institute www.deri.ie
Unified and Native Enrichment
Digital Enterprise Research Institute www.deri.ie
Challenges for Enrichment
1. Determination of Enrichment Source (ES) ¨ E.g. BMS relational database via a connection
string “Server=www.example.com/rdbms;Database=BMS-DB;”.
2. Retrieval of Information Items from the Enrichment Source ¨ E.g. SQL query against a query interface
3. Finding the Complementary Information for an Event in the Enrichment Source ¨ E.g. recursively join on the ID attribute in the
LOCATIONS table and query for n times.
15/30
Digital Enterprise Research Institute www.deri.ie
4. Fusion of Complementary Information with the Event ¨ E.g.
¨ E.g.
Challenges for Enrichment
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(room, “202e”)
(floor, “second floor”)
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(type, "energy consumption”)
(device, "heater x”)
(consumption, "high”)
(location, “202e, second floor”)
16/30
Digital Enterprise Research Institute www.deri.ie
Produce or Consumer-side
n Producer side enrichment ¨ Producers can provide enrichment elements with
events
¨ Identify enrichment source and mechanism
n Consumer side enrichment ¨ Have better knowledge of completeness from
their perspective
¨ Proposed here as a unified element with the subscription matching element
17/30
Digital Enterprise Research Institute www.deri.ie
¨ Enrichment Element – ENRICH FROM identify the enrichment source(s)
– RETRIEVE BY specifies retrieval mechanism for atomic information items
– FIND BY dictate retrieval of information items from the enrichment source(s)
– FUSE BY define fusion approach to integrate retrieved data
¨ Matching Element – as in current matching languages
Unified Subscription
18/30
Digital Enterprise Research Institute www.deri.ie
Formal Model
ES e
MVS(U)
U
HVS(U)
A1 A2
B1
A3
AB1
B2 B3
AB2 AB3
W=e ES∩The universe U, the event e, the enrichment source ES, the world W, the enrichment view HVS, and a matching view MVS
19/30
Digital Enterprise Research Institute www.deri.ie
Formal Model
n Successful Enrichment ¨ Completeness
n Minimal Successfully Enriched Event ¨ Precision
n Approximately Minimal Successfully Enriched Event ¨ Cost to turn an enriched event into a minimal
successfully enriched event
(More details on formal definition in paper)
20/30
Digital Enterprise Research Institute www.deri.ie
Implications
n Sharing & Re-usability of Enrichment Elements ¨ Expert consumers can share enrichment knowledge
with less informed consumers
n Distribution of Enrichment ¨ Distribute enrichment process across nodes to
achieve an optimal overall completeness
n Approximation in Event Processing Engines ¨ Matching over partially complete events would need
to account for still missing information
21 of 30
Digital Enterprise Research Institute www.deri.ie
Linked Data Instantiation
22/30
Digital Enterprise Research Institute www.deri.ie
Linked Data Instantiation
23/30
Event Model Enrichment Source Model Enrichment source is accessible by dereferencing URIs associated with it as per Principles of Linked Data
Digital Enterprise Research Institute www.deri.ie
Linked Data Instantiation
n Unified Subscription Language ENRICH FROM http://www.myenterprise.org/devices
RETRIEVE BY ‘DEREF’
FIND BY ‘Spreading Activation’
‘UniformWeightsAllAdjacent’
FUSE BY ‘UNION’
{?event rdf:type ont:EnergyConsumption.
?event (?p){3} building:SecondFloor.}
24/30
Digital Enterprise Research Institute www.deri.ie
Evaluation Dataset
n English Dbpedia (1st of August 2012) n Constructed from instances of class
dbpedia-owl:Event
n Total of 24,000 events ¨ “Football Match”, “Race”, “Music Festival”, “Space Mission”,
“Election”, “10thcentury BC Conflicts”, “Academic Conferences”, “Aviation Accidents And Incidents In 2001”, etc
n Each event contains one triple in the form: ¨ <eventURI, rdf:type, dbpedia-owl:Event>.
n Enrichment Source: Whole Dbpedia dataset ¨ Potnetially 250 million triples for enrichment
25/30
Digital Enterprise Research Institute www.deri.ie
Evaluation – FIND BY
n FIND BY - Spreading Activation Strategies
n UniformWeightsAllAdjacent ¨ Activation spreads equally to all adjacent nodes
n UniformWeightsRandomAdjacent ¨ Activation spreads equally to random set of
adjacent nodes
n DifferentWeightsSemRel ¨ Activation spreads based on semantic relatedness
to terms in matching element of subscription
26/30
Digital Enterprise Research Institute www.deri.ie
Evaluation – Matching Elements
n Path-shaped graph used to generate matching elements
Matching Element 1 ?event rdf:type dbpedia-owl:Event.
?event (?p){1} bpedia:England_national_football_team
2 ?event rdf:type dbpedia-owl:Event. ?event (?p){2} dbpedia:Queens_Park_Rangers_F.C..
3 ?event rdf:type dbpedia-owl:Event. ?event (?p){3}
dbpedia: Loftus_Road. 4 ?event rdf:type dbpedia-owl:Event.
?event (?p){4} dbpedia: Fulham_F.C..
27/30
Digital Enterprise Research Institute www.deri.ie
n SemRel (unified approach) performs best
n Less effective with complex subscriptions
n Completeness and precision not weighted equally
Enrichment Results
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
1 2 3 4
F5Score
Subscription
UniformWeightsAllAdjacents UniformWeightsRandomAdjacentsDifferentWeightsSemRel
28/30
Digital Enterprise Research Institute www.deri.ie
Summary and Future Directions
29/30
Digital Enterprise Research Institute www.deri.ie
Summary and Future Directions
n Event enrichment can be integrated as a core task of event processing engines ¨ Unified enrichment logic with event subscription
logic
¨ Native enricher tackles incompleteness before matching
n Future Work ¨ Formalizing the enrichment language element
¨ Leverage commonalities among multiple consumers for enrichment optimization
30 of 30
Digital Enterprise Research Institute www.deri.ie
Hasan, S., O’Riain, S., and Curry, E. 2013. “Towards Unified and Native Enrichment in Event Processing Systems,” In 7th ACM International Conference on Distributed Event-Based Systems (DEBS 2013), Arlington, Texas, USA: ACM, pp. 171–182. http://www.edwardcurry.org/publications/DEBS2013_Enrichment.pdf
Further Reading