director, advanced analytics institute university of...
TRANSCRIPT
THINK.CHANGE.DO
Professor Longbing CAO
Director, Advanced Analytics InstituteUniversity of Technology Sydney
Professor Longbing CAO
Director, Advanced Analytics InstituteUniversity of Technology Sydney
Agenda
• Why UI (in the big data era)
• What is UI
• Case studies of UI computing
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational intelligence
Network
IntelligenceWhy Ubiquitous Intelligence is
Becoming Increasingly Important?
Beyond Business Transactions
3 Vs and 3 Is
Big Data Dictionary• Availability , actionable
• Big
• Capacity, complexity, connect , correlate, complete, cost
• Differentiation, diversity
• Exponential
• Faster time, faster response
• Governance, growth, global
• Hierarchy, historical data
• Innovation, intelligence, …
• Linkage
• Multiple
• Precise answers, profit, profitability
• Social media, sensor data, structured, storage issue
• Relevance, relative, relationship
• unknown risks
• unseen patterns, sentiments and relationships
• unsolvable problems
• unstructured, unconnected
• value
Bigger data Bigger business
Reporting � Management � Decision
Business Value-Centred Analysis
Big Data management �
Big Data analytics
Data & Analytics Industry
UI-related Actionable Problem-solving
� Ubiquitous
intelligence is involved
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational intelligence
Network
Intelligence
What is Ubiquitous Intelligence?
Longbing Cao, et al. Ubiquitous Intelligence in Agent Mining.
ADMI 2009: 23-35.
Ubiquitous Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational intelligence
Network
Intelligence
Ubiquitous Intelligence Computing
• Involve, quantify and synthesize Ubiquitous
Intelligence
– What are they?
– How to ‘quantify’ them?
– How to synthesize them for problem-solving?
Data Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
intelligence
Network
Intelligence
Description
• Data Intelligence tells interesting stories and/or indicators hidden in data about a business problem. The intelligence of data emerges in the form of interesting patterns and actionable knowledge.
– General level of data intelligence• Refers to the knowledge identified from explicit data, presenting
general knowledge about a business problem
• Ex. frequent patterns
– Deep level of data intelligence• Refers to the knowledge identified in more complex data, using more
advanced techniques, or disclosing much deeper information and knowledge about a problem.
• Ex. associative classifiers
Types
• Explicit intelligence
– Stock trends
• Implicit intelligence
– Hidden group trading
• Syntactic intelligence
– Itemset associations
• Semantic intelligence
– Casual relation within trading behaviors
Aspects
• Data type such as numeric, categorical, XML, multimedia and composite data
• Data timing such as temporal and sequential
• Data spacing such as spatial and temporal-spatial
• Data speed and mobility such as high frequency, high density, dynamic data and mobile data
• Data dimension such as multi-dimensional, high-dimensional data, and multiple sequences
• Data relation such as multi-relational, linkage record
• Data quality such as missing data, noise, uncertainty, and incompleteness
• Data sensitivity like mixing with sensitive information
• ……
Techniques
• Data quality enhancement
• Data matching and integration
• Information coordination
• Feature extraction
• Parallel computing
• Collective intelligence
• Dimension reduction
• Space mapping
• Computational complexity
• Data privacy and security
• Combined mining
• Visualization
• ……
An example
Behavior Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
intelligence
Network
Intelligence
Description
• Behavior intelligence discloses the process,
impact and utility of a collection of activities
• Individual behavior intelligence
– Buy a stock
• Group behavior intelligence
– Pool manipulation
• Collective behavior intelligence
– Financial crisis
Aspects
• behavior interactions and networks,
• behavioral patterns,
• Behavior anomalies
• behavioral impacts,
• behavior dynamics,
• the formation of behavior-oriented groups and collective intelligence,
• behavioral intelligence emergence etc of human beings, organisms, systems, organizations and artificial entities in conjunction with their environments.
• … …
Techniques
• Behavioral data construction
• Behavioral representation and modeling
• Reasoning about behavior
• Behavioral pattern analysis
• Behavioral impact analysis
• Sequence analysis
• Coupled behavior analysis
• Behavioral intelligence emergence
• Behavior simulation
• Self organizing theory
• … …
Self Organization
• Why does this stock go so crazily?
An example
• Short-term manipulation behaviors as cause
Behavior exterior
presentation
Possible
driver
Possible
behavior
interior
driver
Behaviors of associated accounts as the driver of the price movement
Group
behaviors
Group
behaviors
• What makes multiple objects/behaviors
different?
Key factors:
• Multiple actors
• Multiple behaviors
• Multiple properties
• Coupling relationships
• Organizational factors
How are they handled by existing
techniques?
Time series analysis
Multiple time series analysis
Frequent pattern mining
Sequence analysis
Coupled sequence analysis
Behavior exterior analysis
Behavior interior analysis
Domain Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
intelligence
Network
Intelligence
Description
• Domain intelligence refers to the intelligence
that emerges from the involvement of domain
factors and resources, which wrap not only a
problem but its target data and environment.
• The intelligence of domain is embodied
through the involvement into modeling
process, models and systems.
Types
• Qualitative domain intelligence, refers to the type
of domain intelligence that discloses qualitative
characteristics or involves qualitative aspects.
– Ex. ``beating the market''
• Quantitative domain intelligence, refers to the
type of domain intelligence that discloses
quantitative characteristics or involves
quantitative aspects.
– Ex. ``beat VWAP’’
Aspects
• Domain knowledge,
• Background and prior information,
• Meta-knowledge and meta-data
• Constraints,
• Business process,
• Workflow,
• Benchmarking and criteria definition, and
• Business expectation and interest
• … …
Techniques
• Representation and involvement of domain knowledge
• Ontological engineering and semantic web
• Formal modeling
• Involving domain factors in the data mining model and process
• Interaction design
• Business process and workflow mining
• Constraint-based mining
• Multi-step mining
• Business interestingness
• ……
http://www.ifs.tuwien.ac.at/~lammarsch/HypoVis/abouthypovis.html
An example
• Constrained environment for trading agent studies– Expectation V
– Domain factors• M = {I, A, O, T, R, S, E}
– instruments I={stock, option, future, …}
– participants A={investor, broker, maker, …}
– order O={limit, market, quote, block, stop}
– timeframe T
– market rules R
– trading strategy S
– execution system E
Network Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
intelligence
Network
Intelligence
Description
• Network Intelligence emerges from both web
and broad-based network information, facilities,
services and processing
– The information and facilities from the networks
surrounding the target business problem either
consist of the problem constituents, or can contribute
to useful information and/or support for actionable
knowledge discovery.
Aspects
• Distributed information and resources• Linkages amongst distributed objects• Hidden communities and groups• Information and resources from network and
in particular the web• Networked facilities• Information retrieval• Structuralization and abstraction from
distributed textual (blog) data• ……
Techniques
• Application integration• Information fusion• Feature fusion• Data gateway and management• Distributed computing• Information retrieval and searching• Distributed data mining• Combined mining• Linkage analysis• Group formation• Data mobility• Agent-based distributed/peer-to-peer mining• Agent-based information sharing• ……
An Example
Agent-based multiple data mining
task integration
- One data source:
~ single
~ homogeneous
- Multiple DM task agents:
~ conduct individual tasks
~ coordinate by agent
coordinator and data mining
task coordinator
- Pattern integration:
~ pattern integration agent
~ final integrative patterns
Human Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
intelligence
Network
Intelligence
Description
• Human Intelligence refers to
– explicit or direct involvement of human empirical
knowledge, etc.
• Ex. Tuning parameters via user interfaces
– implicit or indirect involvement of human
intelligence
• Ex. Opinion from an expert group
Aspects
• Human empirical knowledge,• Belief, intention, expectation, • Sentiment, opinion • Run-time supervision, evaluation, • Expert groups• Imaginary thinking, • Emotional intelligence, • Inspiration, • Brainstorm, • Retrospection, • Reasoning inputs, and • Embodied cognition like convergent thinking through
interaction with other members in assessing identified patterns
Savive's Corner:
Mentally Ill in Amityville
(2nd Edition)
Techniques
• Dynamic user modeling
• Online user interaction
• Interaction design
• Group decision-making in pattern discovery
• Adaptive interaction
• Distributed interaction
• Consensus building
• Social computing
• Sentiment analysis
• Opinion mining
• User modeling
• ……
An example
Agent service interfaces:
- Business interfaces supporting business users’ interaction with the system;
- Technical interfaces supporting domain expert’s interaction with the system;
- Algorithm interfaces supporting algorithm designer’s interaction with the system;
- Running interfaces supporting system execution control.
Gerd Doeben-Henisch: Reconstructing Human Intelligence within Computational Sciences
Organizational Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational
Intelligence
Network
Intelligence
Description
• Organizational Intelligence refers to the
intelligence that emerges from involving
organization-oriented factors and resources,
capability for responding to change and
complexity intelligently, leadership, strategy
and environmental conditions.
Aspects
• Organization reality, needs and constraints
• Leadership, strategy and environmental conditions
• Organizational structures
• Organizational behaviors
• Organizational evolution and dynamics
• Organizational/business regulation and convention
• Organizational goals
• Organizational norms and policies
• Organizational actors and roles
• Organizational interaction
• Capability for responding to change and complexity intelligently
• Impact of organizational interaction and dynamics
• Changes
• … …
Techniques
• Organizational computing
• Organizational theory
• Organizational behavior study
• Computer simulation
• Complexity theory
• Divergence and convergence of thinking
• Agent mining
• … …
An Example
Social Intelligence
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Organizational
intelligence
Social
Intelligence
Network
Intelligence
Description
• Social Intelligence refers to the intelligence
that emerges from the group interactions,
behaviors and corresponding regulation
during a process.
– Human social intelligence
– Animate/agent-based social intelligence
• Human social intelligence aspects
– Social cognition,
• Company X is with good fundamental performance
– Emotional intelligence,
• Most traders believe the market downturn is close to over, and
enter the market (from opinion mining)
– Consensus construction, and
• Leadership-driven actions
– Group decision.
• Let’s wait for a while
Techniques
• Social computing
• Social network analysis
• Software agent technology
• Swarm and collective intelligence
• Divergence and convergence of thinking
• Agent mining integration
• Complex systems
• ……
An Example
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational intelligence
Network
IntelligenceCase Studies of UI Modeling,
Analysis and Mining
Metasynthesis of UI
• Surround data, business, environment and
decision
• Involve all relevant UI
• Incorporate into the process, modeling and
systems of solutions
System Architecture to Synthesize UI
Longbing Cao, et al. Combined Mining: Discovering Informative Knowledge
in Complex Data, IEEE Trans. SMC Part B, 41(3): 699 - 712, 2011.
UI Evaluation Framework
Business
Technical
biz_obj() biz_subj()
tech_obj() General
technical
tech_subj() Domain-specific
user-specific
social
Longbing Cao, et al. The evolution of KDD: Towards domain-driven data
mining. International Journal of Pattern Recognition and Artificial
Intelligence, 21(4): 677-692, 2007.
Social Behavior Modeling
Can Wang, Longbing Cao. Modeling and Checking of Complex Agent Behaviors,
Behavior Computing, Springer, 2012
71
Behavior Visual DescriptorIntra-relationships within Behavior
Relationship crossing behaviors
Coupling relationships
• From temporal aspect
• From inferential aspect
• From combinational aspect
21/6/2010 L. Cao 74
Relationship Sub-model
Behavior Ontology Model
2012/11/29 75
Behavior Formal Descriptor
76
Model Transformation
=
� Single Behaviors
Agent behaviors are represented as TSs.
77
� Multiple Behaviors
Formal behavior representation
Behavior checking
21/6/2010 L. Cao 80
Graphical Action Sub-model of Online Shoppingbased on Stages
{ }{ { }},RP AID AP RK→ ⊕○ ○
Coupled Object Analysis
-- Non-iidness Learning
Longbing Cao. Combined Mining: Analyzing Object and Pattern Relations for Discovering and Constructing Complex but
Actionable Patterns, WIREs Data Mining and Knowledge Discovery, 2012.
Can Wang, Zhong She, Longbing Cao. Coupled Clustering Ensemble: Incorporating Coupling Relationships Both between
Base Clusterings and Objects, ICDE2013.
Yin Song, Longbing Cao, et al. Coupled Behavior Analysis for Capturing Coupling Relationships in Group-based Market
Manipulation, KDD 2012, 976-984.
Longbing Cao, et al. Coupled Behavior Analysis with Applications, IEEE Trans. on Knowledge and Data Engineering, 24(8):
1378-1392 (2012).
Can Wang, Longbing Cao, Minchun Wang, Jinjiu Li, Wei Wei, Yuming Ou. Coupled Nominal Similarity in Unsupervised
Learning, CIKM 2011, 973-978.
Longbing Cao, et al. Detecting Abnormal Coupled Sequences and Sequence Changes in Group-based Manipulative Trading
Behaviors, KDD2010, 85-94.
Quelea vs Elephant
Coupling of
intelligence
Current KDD/ML Focus - IIDNESSTraders are
independent
Behaviors of a
trader are treated
independent or
loosely dependent
Associations &
frequent patterns Clustering
Classification
Foundation:-Individual objects/behaviors
-Without coupling relationships
(dependency) between
objects/behaviors
-Focus on local features within
an object/behavior
Clustering
Objective functions:
-K-means
-FCM
Note:
- Xj Individual
objects only!
Question:
- How about Xj1 and
Xj2 dependent?
Iidness Learning
Dependence and similarity between objects matter
Iidness learning
noniidness learning
Coupled Behavior Analysis
2012/11/29
Coupled Nominal Similarity
Intra-coupled Interaction:Inter-coupled Interaction:
88
Application
High Impact Behaviour Analysis
in Social Security
Longbing Cao. Zhao Y., Zhang, C. Mining Impact-Targeted Activity Patterns
in Imbalanced Data, IEEE Trans. on Knowledge and Data Engineering,
20(8): 1053-1066, 2008.
Data intelligence aspects
– Itemset imbalance
– Impact imbalance
– Seasonal effect
– Demographic data
– Transactional data
– Behavioral data
– Policy data
– Interactions
� Itemset/tuple selection/construction
Domain intelligence aspects
– Business process/event for activity selection
– Domain knowledge
– Feature selection
– Sequence construction
– Impact target
• Positive impact
• Negative impact
• Multi-level impacts
�Feature/attribute selection
�Interestingness definition
�New pattern structures
Organizational/social factors
– Operational/intervention activities
– Seasonal business requirement/interaction changes
– Business cost
• debt amount/duration
• intervention cost
– Business benefit
• saving/preventing debt amount or reducing debt duration
• productivity gain
– Deliverable format
Combined Mining
• ‘combined’ method:
– multi-source combined mining (D)
– Multi-feature combined mining (F)
– Multi-method combined mining (R)
Combined Mining
• Impact-oriented pair pattern
• Impact-oriented cluster patterns
High impact behaviour analysis(Impact-targeted behavior pattern mining)
Demographics + behavior mining
Policy 1 Policy 2
Demographic 1 Low value High value
Demographic 2 High value Low value
High Impact Behavior
Group Behaviour Analysis
in Financial Markets
Yin Song, Longbing Cao, et al. Coupled Behavior Analysis for Capturing Coupling Relationships in
Group-based Market Manipulation, KDD 2012, 976-984.
Yin Song and Longbing Cao. Graph-based Coupled Behavior Analysis: A Case Study on Detecting
Collaborative Manipulations in Stock Markets, IJCNN 2012, 1-8.
Longbing Cao, Yuming Ou, Philip S Yu. Coupled Behavior Analysis with Applications, IEEE Trans.
on Knowledge and Data Engineering, 24(8): 1378-1392 (2012).
Longbing Cao, Yuming Ou, Philip S YU, Gang Wei. Detecting Abnormal Coupled Sequences and
Sequence Changes in Group-based Manipulative Trading Behaviors, KDD2010, 85-94.
• Short-term manipulation behaviors as cause
Behavior exterior
presentation
Possible
driver
Possible
behavior
interior
driver
Pool manipulation
• Data structure 2:
CHMM Based Coupled Sequence
Modeling
• Coupled behavior sequences
– Multiple sequences
– Coupling relationship
– Behavior properties
CBA - CHMM
• Benchmark Models
– HMM-B: Buy-based HMM
– HMM-S: Sell-based HMM
– HMM-T: Trade-based HMM
– IHMM: HMM-B + HMM-S + HMM-T
– CHMM: CHMM(buy, sell, trade)
– ACHMM: Adaptive CHMM(buy, sell, trade)
• Business Performance
Learning Analytics for Students
at Academic Risk
Student progression analysisS1
123
4114
03Aut
03Spr
04Aut
04Spr
12
2
102
5
2
131
1 23
05Aut
05Spr
06Aut
32
2
1
06Spr
1
1
2
1
63
07Aut
07Spr
08Aut
08Spr
09Aut
1
8
1
2
2
1
1
2
1
2
2
1
1
S2
S3
142
Contrast Analysis
• Some conclusions
– There is no single discriminative factor associated with student’s
academic performance
– Combination of factors contribute more significantly
48510 factor combination pass fail Rule type
1<activity_num<3 AND 90<TES_score<100 AND
0.193<birth_cntry_fail_rate<0.23352 7 Positive
3<gpa_val<4 AND exclusion_sanction='N' AND
0.19<ctzn_cntry_fail_rate<0.245 31 1 Positive
HSC_physics>80 AND 0.19<ctzn_cntry_fail_rate<0.245 AND
exclusion_sanction='N' 46 6 Positive
activity_num<=1 AND 0.8<gpa_val<1.2 3 61 Negative
0.8<gpa_val<1.2 AND 0.19<ctzn_cntry_fail_rate<0.245 4 63Negative
stu_grade =1 AND pre_mark_33130<50 3 34 Negative
Combined Contrast Analysis
• Conbination of factors makes rules more accurate
48510 factor combinationpass fail
Contrast
Rate
Rule
type
HSC_physics >80 46 6 7.67 Positive
TES_score >90 62 8 7.75 Positive
HSC_physics >80 AND TES_score >90 30 1 30 Positive
0.8<GPA<=1.2 4 75 18.75 Negative
pre_mark_33130<50 13 66 5.07Negative
0.8<GPA<=1.2 AND pre_mark_33130<50 0 32 +infinite Negative
Data
Intelligence
Behavior
Intelligence
Human
Intelligence
Domain
Intelligence
Social
Intelligence
Organizational intelligence
Network
Intelligence
Challenges and Prospects
Challenges of UI Computing
Representation Analysis Mining Learning Reasoning Checking Evaluation
The Trends
Accessibility
Visibility
Operability
CONCLUSIONS
• Increasing demand of real-life problem-solving relies
on a deep understanding of ubiquitous intelligence
• Modeling and analysis of UI is crucial and challenging
• Unlimited opportunities in UI Computing
Your Feedback Is Appreciated!
• Longbing Cao
www-staff.it.uts.edu.au/~lbcao
• UTS Advanced Analytics Institute
www.analytics.uts.edu.au