Title: Web Intelligence Meets BrainInformatics an impending revolution in WI and Brain Sciences
1Web Intelligence Meets Brain-Informatics- an
impending revolution in WI and Brain Sciences -
Ning Zhong
Web Intelligence Consortium Maebashi Institute
of Technology, Japan
2Acknowledgements
- Prof. Piotr Szczepaniak, organizers of AWIC 2005
- Jiming Liu, Yiyu Yao, Jinglong Wu, Edward
Feigenbaum, John McCarthy, Tom Mitchell, Setsuo
Ohsuga, Benjamin Wah, Philip Yu, Lotfi
Zadeh, Xindong Wu etc. - WIC Technical Committee and WIC Research Centers
in Australia, Beijing, Canada, Hong Kong, India,
Japan, Korea, Mexico, Poland, Spain, France, and
UK, among others - Students and Post-doc/Visitors at
- Maebashi Institute of Technology
- Beijing University of Technology
3AIIT Web Intelligence (WI)?
- WI explores the fundamental roles as well as
practical impacts of - Artificial Intelligence (AI) (e.g., knowledge
representation, planning, knowledge discovery,
agents, and social intelligence) and - Advanced Information Technology (IT) (e.g.,
wireless networks, ubiquitous devices, social
networks, and data/knowledge grids) - on the next generation of Web-empowered
systems, services, environments, and activities.
Zhong, N , Liu, J , and Yao, Y Y (eds.) Web
Intelligence, Springer, 2003
4Perspectives of WI (Yao-Zhong-Liu,
2001)
- WI may be viewed as applying results from
existing disciplines (IT and AI) to a totally new
domain. - WI introduces new problems and challenges to the
established disciplines. - WI may be viewed as an enhancement or an
extension of IT and AI.
5A New Perspective of WI WI Meets
Brain-Informatics (BI)
- New instrumentation (fMRI etc) and advanced IT
are causing an impending revolution in WI and
Brain Sciences (BS). - The WI based portal techniques will provide a
new powerful platform for BS. - The new understanding and discovery of human
intelligence models in BS will yield a
new generation of WI research/development.
6A New Perspective of WI
- New instrumentation (fMRI etc) and advanced IT
are causing an impending revolution in WI and
Brain Sciences (BS). - The WI based portal techniques will provide a
new powerful platform for BS. - The new understanding and discovery of human
intelligence models in BS will yield a
new generation of WI research/development.
7What Are Web-Based Portals?
- Enabling an organization (company, community)
to create a virtual organization on the Web where
key production/information steps are outsourced
to partners and customers. - A single gateway to personalized information
needed to enable informed interdisciplinary
research, services, and/or business activities. - One of the most sophisticated applications in
WI.
8Physical World
Virtual World
Knowledge
- Applications
- e/m-Business
- e/m-Science
- e/m-Learning
- e/m-Government
-
- Intelligent INFORMATION Analysis
Internet
Web-Based Information
Web Accessible Information
Intelligent DATA Pre-processing
- Ubiquitous Agents
- Adaptive Interface
- Personal Assistants
- Distributed Cooperative Work
Heterogeneous Information Sources
Intelligent Resource Intermediary Coordination
9What Is Brain-Informatics (BI)?
- Brain Informatics (BI) is a new interdisciplinary
field to study human information processing
mechanism systematically from both macro and
micro points of view by cooperatively using - -- experimental brain/cognitive technology
and - -- WI centric advanced information
technology. - In particular, it attempts to understand human
intelligence in depth, towards a holistic view at
a long-term, global field of vision to understand
the principle, models and mechanisms of human
multi-perception, language, memory, reasoning and
inference, problem-solving, learning, discovery
and creativity.
10Brain Sciences v.s. Brain Informatics
- Although brain sciences have been studied from
different disciplines such as cognitive science
and neuroscience, Brain Informatics (BI)
represents a potentially revolutionary shift in
the way that research is undertaken. - It attempts to capture new forms of collaborative
and interdisciplinary work. - In this vision, new kinds of BI methods and
global research communities will emerge, through
infrastructure on the Wisdom Web and Knowledge
Grids that enables high-speed and distributed,
large-scale analysis and computations, and
radically new ways of sharing data/knowledge
repositories.
11WI centric IT
Brain Informatics
Cognitive Science
Neuroscience
12Our Recent Works (Zhong-Wu, 01-04)
- Investigating human multi-perception mechanism by
cooperatively using cognitive technology and data
mining techniques for developing artificial
systems which match human ability in specific
aspects. - Building a multi-database mining system on
the Wisdom Web and Grid platform for developing a
brain-informatics portal.
13Main Questions
- How to design the psychological and physiological
experiments for obtaining various data from human
multi-perception mechanism? - How to analyze such data from multiple aspects
for discovering new models of human
multi-perception?
14physiological experiments
traditional psychometrics
(fMRI, EEG, )
results (discrete)
results (image)
results (wave)
data modeling and transformation
image data
wave data
discrete data
data mining knowledge discovery
models of human multi-perception
test of the multi-perception models
application of the models
15Data from Physiometry
fMRI experimental apparatus
image of the electroencephalograph (EEG)
The future of cognitive science and neuroscience
will be affected by the ability to do large-scale
analysis of fMRI brain activations.
fMRI functional Magnetic Resonance Imaging
16Design of Experiments
- Human subjects perform simple arithmetic
stimulus
task
17Time Chart
start
end
on-task
off-task
no-task
no-task
5060
180
1015
180
(sec)
(sec)
(sec)
35
18fMRI Images
- Understanding human multi-perception mechanism
by analyzing fMRI images obtained from human
visual and auditory psychological experiments
19Visual and Auditory Calculation
The results of the visual and auditory
calculation processing in fMRI experiments
20Peculiarity Oriented Analysis
- We observe that fMRI/EEG brain activations are
peculiar ones with respect to the specific state
and the related part activated by a stimulus. - We proposed a way of peculiarity oriented mining
(POM) in multiple human brain data. - Peculiarity is a kind of interestingness.
Peculiarity relationships/rules (with common
sense) may be hidden in a relatively small number
of data.
21Peculiarity Oriented Mining (1) (Zhong et al
99-04)
- The main task of peculiarity oriented mining
(POM) is the identification of peculiar data. - Peculiarity Factor (PF) evaluates whether
- occurs in relatively small number and is
very different from other data
22Peculiarity Oriented Mining (2)
23Peculiarity Oriented Mining (3)
- After the PF evaluation, the peculiar data are
selected by using a threshold value -
- where can be specified by a user. That is, if
is over the threshold value,
- is a peculiar data.
24fMRI Data Is Very Large
- The number of the data for each subject is
122,880 (i.e. 64pixelx64pixelx30images) - 9 subjects x 122,880 1,105,920
- A way for data reduction is to use a tool called
Talairach Daemon that is based on the Brodman
human brain map as prior knowledge.
25Brodmann Brain Map (Prior Knowledge)
(a general view of brain functions)
BA
Function
visual information processing auditory
information processing motor area (language
representation) Wernickes area (language
understanding,calculation) Brocas area (language
representation) working memory
17,18,19 22,41,42 6 39,40 (21,22,37) 44,45,47 9,
45,46
26POM Based fMRI Data Mining Processes
formalize/transform fMRI data by MEDx/SPM
obtain BA values based on the TD coordinate
obtain BA values based on the TD coordinate
carry out POM only on the obtained BA values
peculiar data obtained
peculiar data obtained
Data Mining (1)
27A Multi-Step Process
- Various tools (e.g. MEDx/SPM, TD) can be
cooperatively used with POM in a multi-step
process for pre-processing (data cleaning,
modeling and transformation), mining and
post-processing. - Goal Machine-processing/understanding to replace
human-expert centric visualization.
28Part of Results (PF Values) Based on BA (Visual)
29Part of Results (PF Values) Based on BA (Auditory)
30An Interesting Hypothesis
By comparing the results of mining in auditory
and visual calculation, we can confirm an
interesting hypothesis of human information
processing
Auditory information may be transferred into
visual information, in some cases of advanced
information processing such as calculation.
A possible explanation for the hypothesis is that
there may be some common areas of auditory and
visual information mechanisms.
31Observations
- Although many cognitive/brain scientists have
already studied human information processing
mechanism of visual and auditory, separately, the
relevance between visual and auditory information
processing needs to be investigated in depth. - We need to extend multi-data source by including
EEG brain waves for multi-aspect analysis in
various data mining approaches.
32Brain Waves (EEG)
Sample Brain Waves
Visualization in BIMUTAS
- Cooperative use with fMRI images
for multi-aspect analysis
33Comparison Between the Space and Time Resolution
of Various Brain Measurements
- Time Resolution
- 1000s PET
- 100s shape measurement
- blood-flow measurement
- 10s electro-magnetic measurement
-
fMRI OT - 1s unobtrusive
-
MRI - 100ms
- obtrusive
CT - 10m
- MEM OR
MEG EEG - 1ms
- 0.01 0.1 1
10 100 - Space
Resolution mm
fMRI has excellent spatial resolution, EEG has
excellent temporal resolution
34Multiple Data Sources
- Each method (fMRI and EEG) has its own good
points and weakness from the aspects of space and
time resolution. - fMRI provides images of functional brain activity
to observe dynamic activity patterns within
different parts of the brain for a given task. It
is excellent in space resolution, but inferior
time resolution. - EEG provides information about the electrical
fluctuations between neurons that also
characterize brain activity, and measurements of
brain activity at resolutions approaching real
time. - Multiple data sources from various practical
measuring methods are required.
35Position of Electrode
Front
- The number of effective channels 24
- Extension of the international 10-20 system
- Sampling frequency is 500Hz
36A POM Based EEG Data Mining Process(Motomura-Zhon
g, 04-05)
Noise Filtering (FIR Filter)
Format Transformation
Data Extraction
Experiments
Special Format ? Text Data
Check experiment
Model Transformation
POM
Data Consolidation
Evaluating the Mined Results
The Discovered Knowledge
37Preprocessing
- ?The FIR band pass filter (460Hz, 512tap) is
used to remove noises. - ?They are clustered for each task after noise
removal. - ?Data is extracted in consideration of eye
movement.
Invalidation
Available
Blue Beep signal Red Eye movement Black
Brain waves
Blinking
Beep sound
38Model Transformation
- A problem of using the POM in time series data
is that it is influenced by a phase. - That is, it is necessary to shift the cut area
of each data so that correlation may become
strong.
This mining method needs the model transformation
392-variate Histogram
- Transforming the filtered brain waves into
2-variate histogram with the slope and the
potential
Class of the potential (-23.523.5, l45)
Class of the slope (-4.74.7 m47)
Frequency of appearance
40Frequency of appearance is denoted by different
colors or numerical values in cells
Potential mV
62
5862
26
-22
26
-62-58
-62
Slope mV/2msec
11.7
-11.7
0.30.9
-0.30.3
-0.91.5
11.111.7
-11.7-11.1
41Example of 2-variate Histogram
Subject B Position of Electrode
(CH) CZ
42Data Preparation for Findingthe Peculiar Channel
- 1. Forming the data for peculiarity analysis by
calculating the difference between on-task and
no-task in each channel, respectively. - Example
?
on-task
no-task
the data to be used for peculiarity analysis
2. The transformed data are used for POM.
43Finding the Peculiar Channel
44Explanatory Notes
45Experimental Results (1)
46Multi-Aspect Analysis(Zhong-Ohshima, 03-04)
- Using various data mining techniques
(association, peculiarity-oriented,
classification, etc) for analyzing in
multiple data sources. - Studying the neural structures of the activated
areas and try to understand - - how a peculiar part of the brain
operates - - how they work cooperatively to implement
- a whole information processing
- - how they are linked functionally to
individual - differences in performance.
- Changing the perspective of cognitive
scientists from a single type of experimental
data towards a holistic view at a
long-term, global field of vision.
47A Brain-Informatics
Portal Architecture
ontologies
ontologies
ontologies
KDD Services
Data-grid
Data-grid
Data-flow Manager
Mining-flow Manager
Mining-grid
Mining-grid
Knowledge-flow Manager
Knowledge-grid
Knowledge-grid
Grid-Based Middleware
Browsing/Sampling Data KDD Process Planning Model
Analysis/Refinement Visualization
Wisdom Web-2
Private Workspac-1
Private Workspace-2
Coordinating Agent Cooperative
planning Distributing KDD tasks Combining partial
models
Local DB
Local DB
Shared Workspace
KDD process plan
KDD process plan
48Ontologies Based Description/Integration for
Multi-Data Source/DM Agents
- Providing a formal, explicit specification for
integrated - use of multiple data sources in a semantic
way. - Providing conceptual representation about the
sorts and - properties of data/knowledge and DM agents,
as well as relations between data/knowledge and
DM agents. - Providing a vocabulary of terms and relations to
model the domain, and specifying how you view
the data sources and how to use DM agents. - Providing a common understanding of multiple data
sources that can be communicated between
grid-based DM agents.
49An Ontology of Data Mining Agents
Preprocess
Post-process
...
...
...
MVF
...
Cluster
...
50The Wisdom Web for Brain-informatics
(Liu-Zhong-Yao, 02-03)
- To provide
- - not only a medium for information/knowledge
exchange/sharing - - but also a type of man-made resources for
sustainable knowledge creation and scientific
evolution. - The Wisdom Web will reply on grid-like agencies
- - self-organize, learn, and evolve their courses
of actions in order to perform service tasks as
well as their identities and interrelationships
in communities - - cooperate and compete among themselves in order
to optimize their as well as others resources
and utilities.
51WI Technologies for Intelligent Portals
Ubiquitous Computing
Multi-modal Interaction
Web Information Retrieval
i-Portals
Web Mining and Farming
Web Agents
Knowledge Networks and Management
Social Networks
Grid Computing
52WI for i-Portals
Web Intelligence
53Four Levels of WI Support
Application-level ubiquitous computing and social
intelligence utilities
Level-4
Knowledge-level information processing and
management tools
Level-3
Interface-level multi-media presentation standards
Level-2
support
Internet-level communication, infrastructure,
and security protocols
Level-1
54A New Perspective of WI
- New instrumentation (fMRI etc) and advanced IT
are causing an impending revolution in WI and
Brain Sciences (BS). - The WI based portal techniques will provide a
new powerful platform for BS. - The new understanding and discovery of human
intelligence models in BS will yield a
new generation of WI research/development.
55Towards Human-Level WI
- One of fundamental goals of WI research is
to understand and develop Wisdom Web based
intelligent systems that integrate all the
human-level capabilities such as real-time
response, robustness, autonomous interaction with
their environment, communication with natural
language, commonsense reasoning, planning,
learning, discovery and creativity.
56Web Intelligence (WI)
Understanding intelligence in depth
Combining the three intelligence related areas
Brain Sciences
AI
Habituation
Human Intelligence
Machine Intelligence
Social Intelligence
57WI
?Brain Sciences
AI?
Hebbian learning
Spatial representations
Constraint satisfaction
Motivation
Population codes
Probabilistic reasoning
Learning
Habituation
Multi-agent
Multi-perception
Reasoning
Planning
Vision
Emotion
Attention
Knowledge based methods
Language
Memory and forgetting
58WI
?Brain Sciences
AI?
Hebbian learning
Spatial representations
Constraint satisfaction
Motivation
Social networks
Population codes
Probabilistic reasoning
Learning
Community discovery
Small world theory
Habituation
Multi-agent
Multi-perception
Reasoning
Planning
Social dynamics
Social agents
Vision
Emotion
Attention
Knowledge based methods
Language
Memory and forgetting
Groupware
Social Intelligence
59 The Relationship Between WI BI
WI Systems
data sources and requirements
implementation
supports
verification
Brain Informatics
WI Fundamental
research needs
new models
60From fMRI/EEG Experiments to New WI/Cognitive
Models
fMRI
New Cognitive Models
Data Mining/ Reasoning
measuring preprocessing
BI
Knowledge
Imaging
EEG
Data/ Knowledge Grid
Intelligent Infor. Analysis
AIIT
Wave
psychological
New WI Models
Wisdom Web
psychometrics
61The synergy between WI and BI will yield
profound advances in our analyzing and
understanding of the mechanism of data,
knowledge, intelligence and wisdom, as well as
their relationship, organization and creation
process.
- Fundamental and implementation of WI will be
studied as a central topic and in a unique way. - It will fundamentally change the nature of IT in
general and AI in particular. - Towards human-level Web Intelligence
62Our Recent Works
- Semantic Web mining/farming and automatic
construction/management of ontologies - Semantic social networks for intelligent portals
- Modeling/representing/managing user behavior
- PSML and distributed Web inference engine
- Wisdom Web based computing
63Our Recent Works
- Semantic Web mining/farming and automatic
construction of ontologies - Semantic social networks for intelligent portal
- Modeling/representing/managing user behavior
- PSML and distributed Web inference engine
- Wisdom Web based computing
64Existing Web Information Structures
- WI must study both portal-centralized and
distributed information sources. - Web information/knowledge could be
- either globally, distributed throughout the Web
within multi-layer over the infrastructure of Web
protocols, - or locally, portal-centralized Web services
- (i.e. the intelligent service provider) that is
integrated to its own cluster of specialized
intelligent applications.
65Web-based Problem-solving System
- To develop a Web-based problem-solving system for
portal-centralized, query-answering based Web
intelligent services/decision-making. - The core of such a system is the PSML (Problem
Solver Markup Language) and PSML-based
distributed Web inference engines.
66The Main Support Functions
- The expressive power and functional support
(e.g. CSP, agents, combined network
reasoning) in PSML for complex adaptive,
distributed problem solving. - Performing automatic reasoning on the Web by
incorporating globally distributed contents and
meta-knowledge automatically collected and
transformed from the Semantic Web and social
networks with locally operational knowledge/data
bases.
67The Main Support Functions (2)
- Representing and organizing multiple, huge
knowledge/data sources for distributed network
reasoning. - Combining multiple reasoning methods in PSML
representation and distributed inference engines,
efficiently and effectively. - Modeling user behavior and representing/managing
it as a personalized model dynamically.
68Distributed Query-Answering Model
comparing the speed of our own printer with other
companies products
users
query
answer
query decomposition
answer synthesis
sub-answer
sub-query
price
monoSpeed
resolution
...
WIE
WIE
WIE
cooperation
69Distributed Web Structure
problem
Description
useful to the problem
Constraint
Nodei
Query
PSML
dividing the query into sub-queries
Meta-knowledge
classifying queries
Inference Engine
Meta-knowledge
Knowledge Base
Constraint Knowledge
case-based reasoning
Knowledge Grid
Domain Knowledge
Node
solving sub-queries
Su-Zheng-Zhong-Liu et al, 04
70PSML and Web Inference Engine
- A feasible way as a step to implement such
a PSML based Web Inference Engine is to
combine OWL and Horn clauses plus agent
technology (dynamic contents, distributed
representation, meta-knowledge collection,
and transformation agents).
71Web Inference Engine
The Semantic Web Social Networks
Dynamic,Global Sources
MetaDATA Contents
MetaDATA Contents
MetaDATA Contents
OWL/XML
Transformation
Output
KAUS
Decision-making Support
KAUS
KB
DB
KB
DB
KB
DB
KB
DB
coupling
Static,Local Sources
72Web Inference Engine
The Semantic Web Social Networks
Dynamic,Global Sources
MetaDATA Contents
MetaDATA Contents
MetaDATA Contents
OWL/XML
Transformation
Output
KAUS
Decision-making Support
KAUS
KB
DB
KB
DB
KB
DB
KB
DB
coupling
Static,Local Sources
73Transformation Process
(Tomita-Zhong-Yamauchi04)
Multiple Portal Sites
A Printer Ontology (OWL)
Global
Local
Application-1 (unifying expressions)
Semantic unified content info. (RDF Schema)
Application-2 (transforming to KAUS)
Ontology for vocabulary unification (OWL)
KAUS sources
74Web Inference Engine
The Semantic Web Social Networks
Dynamic,Global Sources
MetaDATA Contents
MetaDATA Contents
MetaDATA Contents
OWL/XML
Transformation
Output
KAUS
Decision-making Support
KAUS
KB
KB
DB
DB
KB
DB
KB
DB
coupling
Static,Local Sources
75Next Questions
- How to design fMRI/EEG experiments to understand
the principle of human reasoning and problem
solving in depth? - How to understand and predict user profile
and behavior? - How to implement human-level reasoning and
problem solving on the Web?
76Summary
- WI technologies will produce the new tools and
infrastructure components necessary to create
intelligent portals that serves its users
wisely . - New instrumentation (fMRI etc) and IT are causing
an impending revolution in WI and Brian
Sciences. - The ultimate goal is to establish the foundations
of Web Intelligence by studying Brain
Informatics for developing Wisdom Web based
intelligent systems that integrate all the
human-level capabilities.
77To Learn More about WI
wi-consortium.org
Books
Journals
- The IEEE/WIC/ACM
- International
- Conference on
- Web Intelligence
- AWIC
- IEEE-CS TCII
78(No Transcript)
79The International WIC Institute (WICI)
the right group, at the right time
WIC-Australia Research Centre WIC-Beijing
Research Centre WIC-Canada Research Centre
WIC-France Research Centre WIC-HK Research
Centre WIC-India Research Centre
WIC-Japan Research Centre WIC-Korean Research
Centre WIC-Mexico Research Centre WIC-Poland
Research Centre WIC-Spain Research Centre
WIC-UK Research Centre
WIC
80Thank You !
81References Further Reading
- N. Zhong, J. Liu, Y.Y. Yao (eds.) Web
Intelligence, Springer, March 2003 - J. Liu, N. Zhong, Y.Y. Yao, Z. Ras The Wisdom
Web New Challenges for Web Intelligence (WI),
Journal of Intelligent Information Systems, 20
(1) 2003 - N. Zhong, J. Liu, Y.Y. Yao (eds.) IEEE Computer
SI on Web Intelligence, 35 (11) (2002) - N. Zhong, J. Liu, Y.Y. Yao In Search of The
Wisdom Web, IEEE Computer, 35 (11) (2002) - N. Zhong, Y.Y. Yao, J. Liu, S. Ohsuga (eds.) Web
Intelligence Research and Development, LNAI
2198, Springer (2001) - N. Zhong, J. Wu, A. Nakamaru, M. Ohshima, H.
Mizuhara Peculiarity Oriented fMRI Brain Data
Analysis for Studying Human Multi-Perception
Mechanism, The Cognitive Systems Research
journal, Elsevier (2004) - N. Zhong, Y.Y. Yao, M. Ohshima Peculiarity
Oriented Multi-Database Mining,
IEEE Transactions on Knowledge and Data
Engineering, 15 (4) (2003) - Y.Y. Yao, N. Zhong, J.J. Huang, C.X. Ou, C.N. Liu
Using Market Value Functions for Targeted
Marketing Data Mining, IJPRAI, 16 (8) (2002) - Y. Sai, Y.Y. Yao, N. Zhong Data Analysis and
Mining in Ordered Information Table, Proc.
IEEE ICDM 2001 - N. Zhong, J.Z. Dong, S. Ohsuga Rule Discovery by
Soft Computing Techniques, Neurocomputing, an
international journal, 36 (1-4) (2001) - SI on e-Services, CACM, Vol 46, No 6., June 2003
- SI on Service-Oriented Computing, CACM, Vol 46,
No. 10, Oct 2003 - SI on e-Science, IEEE Intelligent Systems,
Jan/Feb 2004
82References Further Reading (2)
- N. Zhong Towards Web Intelligence, Proc.
AWIC03 LNAI Springer (2003) - Y.Y. Yao, N. Zhong, J. Liu, S. Ohsuga Web
Intelligence (WI) Research Challenges and Trends
in the New Information Age, in N. Zhong et al
(eds) LNAI 2198 (2001) - T. Berners-Lee, J. Hendler, O. Lassila, The
Semantic Web, Scientific American, 284 (5)
(2001) - H.P. Alesso and C.F. Smith, The Intelligent
Wireless Web, Addison-Wesley (2000) - R. Kumar, P. Raghavan, S. Rajagopalan, A. Tomkins
The Web and Social Networks, IEEE Computer
Special Issue on Web Intelligence, 35 (11) (2002) - A. Congiusta, A. Pugliese, D. Talia, P. Trunfio
Designing Grid Services for Distributed
Knowledge Discovery, Web Intelligence and Agent
Systems, an international journal, IOS Press,
1 (2) (2003) - N. Zhong, C. Liu Dynamically Organizing KDD
Processes, IJPRAI, World Scientific (2001) - Y. Sai, Y.Y. Yao, N. Zhong Data Analysis and
Mining in Ordered Information Tables, Proc.
ICDM01 (2001) - Y. Li, N. Zhong An Ontology-Based Web Mining
Model TKDE (submitted) - J.Z. Ji, C. Liu, Z. Sha, N. Zhong Personalized
Recommendation Based on a Multilevel Customer
Model, IJPRAI (submitted) - J. Liu Web intelligence (WI) Some Research
Challenge, Proc. IJCAI03. - J. Liu, S. Zhang, J. Yang Characterizing Web
Usage Regularities with Information Foraging
Agents , TKDE, 16 (5) (2004) - N. Zhong, J. Liu (eds.) Intelligent Technologies
for Information Analysis, Springer (2004)