Intelligent expert systems for location planning

Intelligent Expert Systems
for Location Planning
Daizhong Tang Jiangang Shi and Wei Wang
Nov 2014
Keywords:Expert Systerms,Location Planning, Bayesian Network

Abstract
 Semantic Web technologies can support information
integration and create semantic mashups.
 Web 2.0 enabled contributions to the Web development
on an unprecedented scale
 Through the ontology, the expert system allows
integration of heterogeneous information.
 An intelligent expert system for location planning
 An integrated knowledge process is developed to
guarantee the whole engineering procedure.
 Based on Bayesian network technique, the system
recommends well planed attractions to a user.

5
The Semantic Web
“The Semantic Web is an extension of the
current web in which information is
given well-defined meaning, better
enabling computers and people to
work in co-operation.“
[Berners-Lee et al, 2001]

6
Today’s Web
 Currently most of the Web content is suitable
for human use.
 Typical uses of the Web today are information
seeking, publishing, and using, searching for
people and products, shopping, reviewing
catalogues, etc.
 Dynamic pages generated based on information
from databases but without original information
structure found in databases.

7
Limitations of the Web Search today
 The Web search results are high recall,
low precision.
 Results are highly sensitive to vocabulary.
 Results are single Web pages.
 Most of the publishing contents are not
structured to allow logical reasoning and
query answering.

9
What is a Web of Data?
Thinking back a bit... 1994
HTML and URIs
Markup language and means
for connecting resources
Below the file level
Stopped at the text level
[Miller 04]

10
What is a Web of Data?
(continued)
Now
XML, RDF, OWL and URIs
Markup language and means for
connecting resources
Below the file level
Below the text level
At the data level
[Miller 04]

11
The Syntactic Web
[Hendler & Miller 02]

12
i.e. the Syntactic Web is…
 A place where
 computers do the presentation (easy) and
 people do the linking and interpreting (hard).
 Why not get computers to do more of the
hard work?
[Goble, 03]

14
Web 2.0
 It is all about people, collaboration,
media, ...
[The mind-map pictured above constructed by Markus Angermeier, source Wikipedia]

15
Web 2.0 and Folksonomies
[http://flickr.com/photos/tags/]

16
Distinguishing the meaning
 It is simply difficult for machines to
distinguish the meaning of:
I am a philosopher.
from
I am a philosopher, you may think.
Well,…

17
…Limitations of the Web today
The Web activities are mostly focus on Machine-to-Human,
and Machine-to-Machine activities are not particularly well
supported by software tools.
[Davies, 03]

18
How Can the Current Situation be
Improved?
 An alternative approach is to represent
Web content in a form that is more easily
machine-accessible and to use intelligent
techniques to take advantage of these
presentations.

19
Machine Accessible Meaning
CV
name
education
work
private
[Davies, 03]

20
XML
<H1>Internet and World Wide Web</H1>
<UL>
<LI>Code: G52IWW
<LI>Students: Undergraduate
</UL>
<H1>Internet and World Wide Web</H1>
<UL>
<LI>Code: G52IWW
<LI>Students: Undergraduate
</UL>
HTML:
<module>
<title>Internet and World Wide Web</title>
<code>G52IWW</code>
<students>Undergraduate</students>
</module>
<module>
<title>Internet and World Wide Web</title>
<code>G52IWW</code>
<students>Undergraduate</students>
</module>
XML:
User definable and domain specific markup

21
XML: Document = labeled tree
module
lecturertitle students
name weblink
<module date=“...”>
<title>...</title>
<lecturer>
<name>...</name>
<weblink>...</weblink>
</lecturer>
<students>...</students>
</module>
=
 DTD: describe the grammar and structure of
permissible XML trees
 node = label + contents

22
But What about this?
CV
name
education
work
private
< >
< >
< >
< >
< >
< Χς >
< ναµε >
<εδυχατιον>
<ωορκ>
<πριϖατε>
[Davies, 03]

23
XML
 Meaning of XML-Documents is intuitively clear
 due to "semantic" Mark-Up
 tags are domain-terms
 But, computers do not have intuition
 tag-names do not provide semantics for machines.
 DTDs or XML Schema specify the structure of
documents, not the meaning of the document contents
 XML lacks a semantic model
 has only a "surface model”, i.e. tree

24
XML is a first step
 Semantic markup
 HTML  layout
 XML  content
 Metadata
 within documents, not across documents
 prescriptive, not descriptive
 No commitment on vocabulary and modelling
primitives
 RDF is the next step
[Davies, 03]

25
RDF: Basic Ideas
 Statements
 A statement is an object-attribute-value
triple.
 It consists of a resources, a property, and a
value.
http://mitpress.mit.edu/catalog/item/default.asp?ttype=2&tid=10140
publishedBy
#MIT Press

26
RDF Schema: Basic Ideas
 RDF is a universal language that enables
users to describe their own vocabularies.
 But, RDF does not make assumption about
any particular domain.
 It is up to user to define this in RDF
schema.

27
What does RDF Schema add?
• Defines vocabulary for RDF
• Organizes this vocabulary in a typed hierarchy
• Class, subClassOf, type
• Property, subPropertyOf
• domain, range
AlanTom
Staff
Lecturer Research Assistant
subClassOf
subClassOf
type
supervisedBy
domain range
type
supervisedBy
[adapted from: Studer et al, 04]
Schema(RDFS)
Data(RDF)

28
Basic Queries
 The example provided in RQL.
 Using select-from-where
 select specifies the number and order of
retrieved data.
 from is used to navigate through the data
model.
 where imposes constraints on possible
solutions

29
Basic Queries: Example
select X,Y
From {X} writtenBy {Y}
X, Y are variables, {X} writtenBy {Y}
represents a resource-property-value
triple

31
Ontologies
 The term ontology is originated from
philosophy. In that context it is used as
the name of a subfield of philosophy,
namely, the study of the nature of
existence.
 For the Semantic Web purpose:
 “An ontology is an explicit and formal
specification of a conceptualisation”.
(R. Studer)

32
Ontologies and Semantic Web
 In general, an ontology describes formally a
domain of discourse.
 An ontology consists of a finite list of terms and
the relationships between the terms.
 The terms denote important concepts classes of
objects of the domain.
 For example, in a Tourism, Transportation,
Attraction, Culture, Shopping, General
information, Accommodation, Dinning, and
News & Events are some important concepts.

33
OntologyF-Logic
similar
OntologyF-Logic
similar
PhD StudentDoktoral Student
Object
Person Topic Document
Tel
PhD StudentPhD Student
Semantics
knows described_in
writes
Affiliation
described_in is_about
knowsP writes D is_about T P T
DT T D
Rules
subTopicOf
• Major Paradigms: Logic Programming, Description Logic
• Standards: RDF(S); OWL
ResearcherStudent
instance_of
is_a
is_a
is_a
Affiliation
Affiliation
Siggi
AIFB+49 721 608 6554
A Sample Ontology
[Studer et al, 04]

34
PhD StudentPhD Student AssProfAssProf
AcademicStaffAcademicStaff
rdfs:subClassOfrdfs:subClassOf
cooperate_withcooperate_with
rdfs:range
rdfs:domain
Ontology
<swrc:AssProf rdf:ID="sst">
<swrc:name>Steffen Staab
</swrc:name>
...
</swrc:AssProf>
http://www.aifb.uni-karlsruhe.de/WBS/sst
Anno-
tation
<swrc:PhD_Student rdf:ID="sha">
<swrc:name>Siegfried
Handschuh</swrc:name>
...
</swrc:PhD_Student>
Web
Page
http://www.aifb.uni-karlsruhe.de/WBS/shaURL
<swrc:cooperate_with rdf:resource =
"http://www.aifb.uni-
karlsruhe.de/WBS/sst#sst"/>
instance of
instance
of
Cooperate_with
Ontology & Annotation
Links have explicit meanings!
[Studer et al, 04]

35
Ontologies (OWL)
 RDFS is useful, but does not solve all possible
requirements
 Complex applications may want more possibilities:
 similarity and/or differences of terms (properties or classes)
 construct classes, not just name them
 can a program reason about some terms? E.g.:
 “if «Person» resources «A» and «B» have the same «foaf:email»
property, then «A» and «B» are identical”
 etc.
 This lead to the development of OWL (Web Ontology
Language)
source: Introduction to the Semantic Web, Ivan Herman, W3C

36
Ontology Languages for the Web
 RDF Schema is a vocabulary description
language for describing properties and
classes of RDF resources, with a
semantics for generalization hierarchies
of such properties and classes.
 OWL is a richer vocabulary description
language for describing properties and
classes.

37
Classes in OWL
 In RDFS, you can subclass existing
classes… that’s all.
 In OWL, you can construct classes from
existing ones:
 enumerate its content
 through intersection, union, complement
 through property restrictions
source: Introduction to the Semantic Web, Ivan Herman, W3C

39
Web Services
 Web Services provide data and services to other
applications.
 Thee applications access Web Services via
standard Web Formats (HTTP, HTML, XML, and
SOAP), with no need to know how the Web
Service itself is implemented.
 You can imagine a web service like a remote
procedure call (RPC) which it returns a
message in an XML format.

40
The Promise of Web Services
[Stollberg et al., 05]

41
Semantic Web Technology
+
Web Service Technology
Semantic Web Services
=> Semantic Web Services as integrated solution for
realizing the vision of the next generation of the Web
• allow machine supported data interpretation
• ontologies as data model
automated discovery, selection, composition,
and web-based execution of services
[Stollberg et al., 05]

43
Why the Excitement?
 What are they?
 Bayesian nets are a network-based framework for representing and
analyzing models involving uncertainty
 What are they used for?
 Intelligent decision aids, data fusion, feature recognition, intelligent
diagnostic aids, automated free text understanding, data mining
 Where did they come from?
 Cross fertilization of ideas between the artificial intelligence, decision
analysis, and statistic communities
 Why the sudden interest?
 Development of propagation algorithms followed by availability of easy
to use commercial software
 Growing number of creative applications
 How are they different from other knowledge representation and
probabilistic analysis tools?
 uncertainty is handled in mathematically rigorous yet efficient and
simple way
 representation of problems, use of Bayesian statistics, and the synergy
between these

44
Bayes Rule
 Based on definition of conditional probability
 p(Ai|E) is posterior probability given evidence E
 p(Ai) is the prior probability
 P(E|Ai) is the likelihood of the evidence given Ai
 p(E) is the preposterior probability of the evidence
A1
A2 A3 A4
A5A6
E
∑∑

45
BN Software
 Protégé addin for BN’s Export
 Netica have a API for use BN in another
Applications (have demo)
 GENIE on SMILE is opensource
 100+ Program is developed with source and
without source can use in projects

47
Semantic Web & Knowledge
Management
 Organising knowledge in conceptual
spaces according to its meaning.
 Enabling automated tools to check for
inconsistencies and extracting new
knowledge.
 Replacing query-based search with query
answering.
 Defining who may view certain parts of
information

48
Knowledge Engineering Process
 These stages are done iteratively
 Stops when further expert input is no
longer cost effective
 Process is difficult and time consuming
 As yet, not well integrated with methods
and tools developed by the Intelligent
Decision Support community
[S.O. Rezend et al.,2000 WIT Press]

49
Knowledge discovery
 There is much interest in automated
methods for learning BNS from data
 parameters, structure (causal discovery)
 Computationally complex problem, so
current methods have practical
limitations
 e.g. limit number of states, require variable
ordering constraints, do not specify all arc
directions
 Evaluation methods

50
The knowledge engineering
process
 1. Building the BN
 variables, structure, parameters, preferences
 combination of expert elicitation and knowledge
discovery
 2. Validation/Evaluation
 case-based, sensitivity analysis, accuracy testing
 3. Field Testing
 alpha/beta testing, acceptance testing
 4. Industrial Use
 collection of statistics
 5. Refinement
 Updating procedures, regression testing

52
Overview of the system
ÉZÀ »ª Ì^˜e É˜Ì]Ä°^˜µÓ|f˜YÂmÁ˜ã¶»Z˜
ÉZÆfËZ‡[Á
É‡Ë‡Ä»Z¿‡]
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
dÌÆm‡YcZ‡z‡»
Ã{Y{‡]Y
‡À¯Y‡ecZ‡z‡»
Locationonto
Ê˜ZÀ˜Êf˜Å½Z¼fyZ˜
 divided into 3 parts
 Firstly ,is the
metadata consisting of
preference profile and
transaction profile.
 Secondly, the
information repository
of Ontology was build.
 Finally, the interface
is the part for user
query.
25872 pages(last)
105809 pages(today)

53
Ontology Building
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
dÌÆm‡Yc Z‡z‡»
Ã{Y{‡]Y
‡À¯Y‡ec Z‡z‡»
Locationonto
 Ontology is the central mechanism of the
system.
 Tourist information and service resources are
classiﬁed according to a common ontology
 As currently there is no existing commonly adopted
ontology for tourism.
 needs the expertise of experienced tourist
consultants
 Once the ontology framework is established,
automated method could be used to replace human
effort

54
Ontology engineering
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto
 more than 80 websites for location planning in
China
 Web developers often group related contents
into categories.
 collected a number of websites Yahoo!
Directory. After removing duplicates, we were
left with 232 websites.
 ﬁltering and grouping similar terms to create
upper level ontology.

55
Ontology engineering
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto
 After collecting and analyzing :
 Web Ontology Language (OWL)
 recommended by the World Wide
Web Consortium (W3C)
 used to represent the ontology due to
its capability of explicitly representing
the concepts and their relationships.
 The travel ontology is
modeled using Protege
 Protégé is a free, open-source
platform with
 a friendly user interface that provides
a set of tools

56
Tourism.OWL
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto

57
Estimating user preferences
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto
 After ontology building is performed,we can
construct a Bayesian network for modelling users
preferences.
 nodes selection
 topology building
 parameters setting
predeﬁned regarding the ontologies
Survey or Learn

58
Qualitative, Quantitative
and updating stage ÉZÀ »ª Ì^˜e É˜Ì]Ä°^˜µÓ|f˜YÂmÁ˜˜a¶»Z˜
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto
 the relevant variables are deﬁned
 relationships between the variables have to be
established
 Bayesian Network is Published
 probability distributions assign
 conditional probability tables set
 using Bayes theorem
 Update data

59
Integrated Bayesian
Knowledge
Engineering Process ÉZÀ »ª Ì^˜e É˜Ì]Ä°^˜µÓ|f˜YÂmÁ˜˜a¶»Z˜
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto
 a spiral engineering process is necessary
 developed an integrated Bayesian knowledge
engineering process (I-BKEP)

60
Integrated Bayesian
Knowledge
Engineering Process ÉZÀ »ª Ì^˜e É˜Ì]Ä°^˜µÓ|f˜YÂmÁ˜˜a¶»Z˜
ÉZÆfËZ‡[Á
‡Z]cZ‡Ô‡Y
Google Map
Yahoo Weather
Wiki information
˜ÂeÂ»mashup
É˜]˜Z¯˜]Y˜
Ã{Y{‡]Y
Ã{Y{‡]Y
Locationonto

61
Sensitive Analysis Methods
for I-BKEP
 Sensitivity analysis can be quantiﬁed
using two types of measures:
 entropy : used to evaluate the uncertainty or
randomness of a variable X characterized by a
probability distribution
 mutual information : measures the amount of
information one random variable contains about
another

63
System implementation
 Ruby Programming Language
 Rails addin for web developing
 ROR framework
 Ruby Meta-Programming support
 Simple use of a Web services(REST,RSS,Atom)
 Three-tiered way implementation
 Standard Web site
 Two Type of server : application server and
the Web server
 spatial Web services

64
prototype of this system
 Apache Server
 Ruby on Rails and Google Map API
 OpenStreetMap Code
 The Netica API Programmers Library, is
embedded in the Web server side to estimate a
travelers preferences in a Bayesian network.
 Integrated information about tourist
attractions is represented in OWL

65
Scenario
 Eric is living in New York and he wants to go to
Tongji university by airplane today to attend a
academic conference. We know that his
preference are horse riding, golf, swimming in
order from proﬁle. However, it might be rainy
in Shanghai. Please make a location planning
for him.

66
Scenario in OWL
 Result of OWL Select Query:

67
Sensitivity analysis For Each BN’s Node
 personality and
motivation are the
variables having the
greatest inﬂuence
on the whole
network.
 analysis results can
be checked by
domain expert
(agreed)and utilized
in the next iteration.

69
Intelligent Expert Systems for Location
Planning Based on Smart Phone Data
 increase of mobile network users, the
development of mobile networks and the
occurrence of new information service
 As for as customers and users, they have
different demands to information due to
different interests, different purpose and
different environment.
 A good number of works have been conducted
in location based services domain for addressing
various problems

70
 mining user's interest can find users
behavior on smart phone data such as:
 Location roaming
 Searching in search engines
 Installing applications
 Browsing in Browsers
 Using social networks

71
 Add accuracy to location planning with
adding Probability of User interesting in
Bayesian network and set this value with
users daily behavior.
 Find user interest with k-center
algorithm.

72
 Building mobile user's interests model
 1) Analyze user's logs, include detailed information
about downloading news, the mobile user's location
and the URLs that the user downloads.
 2) Compute the user's interests degree to every
category in a day
 3) Produce the matrix of user's interests: is defined
as a matrix which is made of k rows and d lists
 4) Compute the user's base interest degree on the
kth category
 5) Arraign the categories according to the results
calculated above

73
 User interest(Based ontology) categories:
 Social
 Sports
 Estate
 Movie
 International(culture)
 Technology and Science
 Games
 Politic

74
 Implementation:
 Android OS Platform(85% global market)
 OpenStreetMap Foundation(opensource,Bing)
 Netica API(Bayesian Network using in server)
 PHP(Web 2.0 server)
 REST,JSON,API,OWL,XML Compatible
 Apache Web server
 Protégé(Ontology Development)

Intelligent expert systems for location planning

More Related Content

What's hot (20)

Similar to Intelligent expert systems for location planning (20)

Recently uploaded (20)

Intelligent expert systems for location planning