Information Agents

1
Information Agents( the Semantic Web)

• Martin Beer,
• School of Computing Management Sciences,
• Sheffield Hallam University, Sheffield,
• United Kingdom
• m.beer_at_shu.ac.uk

2
Outline

• Semantic Web How can it help?
• B2B Perspective
• m-Commerce Perspective
• Information Agents A way to Share Information
  in an organization
• Ideas behind Semantic Web are not all new
• What are they
• An Example Application BT Jasper

3
Vision

• The Webs Paradox
• Power Massive amount of content
• Weakness Inability to harness all this content
• Vision Make Web content and services
  machine-understandable
• Support significantly higher levels of automation
• Agents and other intelligent technologies

4
Syntactic Approaches Wont Do

• Domain-specific syntactic tags
• All parties have to agree upfront on a common
  terminology
• Makes it difficult to introduce new terms and
  concepts
• Or modify the meaning of some terms
• Doesnt necessarily guarantee that everyone has
  the exact same understanding of what each concept
  means

5
Semantic Web

• So, lets use ontologies and semantic markup
• But
• How will we get people/organizations to annotate
  Web content?
• The old chicken egg problemb
• Easy-to-use editing tools
• And how will we solve the many other technical
  problems that need to be addressed?
• e.g. ontology translation/equivalence, etc.

6
The Opportunity is Right Now!

• Creating compelling services and functionality is
  the only way to bring industry onboard
• Government programs in the EU and the US have
  agreed to provide around / 50-100M in seed
  money
• The window of opportunity is shortperhaps 3 or 4
  years
• Focus on the easy pickings first!

7
Some Promising Areas

• B2B Interoperability
• Mobile Internet Services
• and many more
• B2C, e.g.
• One-stop online travel services
• Customer protection online dispute resolution
• Medical domain

8
The Emerging Internet Economy
Source Gartner March 2001 Includes sales of
all goods and services for which the order taking
process was completed via the internet - i.e.
excludes proprietary networks
9
Dynamic Supply Chains
Functional Silos
Inventory Control
Purchasing
Production
Sales
Distribution
Manufacturing Management
Distribution
Customers
Materials Management
Suppliers
Enterprise Integration
Suppliers
Internal Supply Chain
Customers
Supply Chain Integration
e-Markets/Exchanges
Buyers/ Sellers
Buyers/ Sellers
Dynamic Internet-enabled Supply Chain
Buyers/ Sellers
Buyers/ Sellers
10
Beyond Just Procurement

• All activities will increasingly be carried out
  across dynamic webs of companies
• Design, production, distribution, maintenance,
  etc.
• Inter-enterprise collaboration
• Dynamic partnerships
• Objective Always work with the best partner
• A companys competitiveness is determined by its
  ability to interoperate with others
• Semantic B2B interoperability

11
Semantic B2B An Example
Car Seat Manufacturers
Electronics Manufacturers
Quote Sensor Design
Exchange
e-Service Center
Collaborative Design
Event Description
Collaborative Diagnosis
Specs Price Reqts.
Diagnosis of airbag that incidentally inflated
12
Beyond Inter-Enterprise Collaboration

• Disaggregate large enterprise solutions
• e.g. large ERP solutions
• Towards interoperation of best-of-breed modules
• Both static dynamic models
• e.g. Dynamic ASP model based on semantic markup
• Potential Benefits
• Lower costs
• More competition You only buy what you need
• Best functionality
• Lower consulting fees?
• Integration, maintenance, etc.

13
Dynamic B2B Interoperability

• Company services and products described with
  semantic annotations
• Services can be combined in an arbitrary fashion
  subject to semantic service descriptions
• e.g. Use Company Xs CAD tool, Company Ys
  manufacturing facility, Company Zs logistics
  system, etc.
• Companies advertise their services in directories
  and/or e-marketplaces just like they advertise
  their products today

14
Emerging Vision

• Semantic Markup
• Service Capability
• Rate
• etc.

Supporting ASP
Supporting ASP
Market-driven Partnership
Core Business Partner
Core Business Partner
Supporting ASP
Core Business Partner
Supporting ASP
Supporting ASP
15
Mobile Internet Services

• The emerging mobile internet
• Towards a billion mobile phone users
• Also PDAs, pagers, wearable computers, etc.
• Most devices to become internet-enabled within a
  few years
• More accurate location tracking functionality to
  become widespread

16
Context Awareness

• Device limitations
• Time critical nature of many usage scenarios
• Require personalization context-awareness

17
Challenges

• Capture user context while minimizing user input
• Match users context with available services
  (push and pull)
• Be useful rather than annoying
• Scale across a broad range of services
• interoperability
• Capture users permission/privacy requirements
• Including sharing of contextual information
• With whom, under which conditions, etc.
• User acceptance is the ultimate criterion

18
Context-Aware Campus Services

• Motivation
• Campus as everyday life microcosm
• Objective
• Enhance campus life through context-aware
  services accessible over the WLAN
• Approach
• Involve stakeholders in the design (e.g.
  students)
• Exploit location, calendar and other sources of
  contextual information
• Develop evaluate ontologies, incl. permission
  profiles
• Evaluate overall acceptance extrapolate

19
Carnegie Mellons Mobile Context-Aware Campus
Services
In collaboration with the Aura consortium
20
Example A Calendar Ontology

• Taxonomy of Activities
• Attending class, studying, taking an exam,
  socializing, etc.
• Actors
• Self, classmates, teacher, etc.
• Permissions Default Preferences
• e.g. when in class, I dont like to be disrupted
  by promotional messages
• which can be selectively overridden by the user

21
Agent-Based Matchmaking

• Matches users contexts and services
• Both push and pull scenarios
• Push scenarios subject to permission profile as
  defined in the users current context
• Pull Queries are customized based on the users
  current context

22
How Internet Agents Work

• The services proposed are not new
• They are already provided (in parts) by Internet
  Agents
• typically embedded within an internet browser
• use a host of internet management tools such as
  Spiders and search engines to gather information

23
How Internet Agents Work
24
The Internet Softbot(Etzioni Weld 1994)

• user makes a high-level menu-based request e.g.
  send the budget memos to Mitchell at CMU
• softbot uses search and inference knowledge to
  determine how to satisfy the request in the
  internet
• softbot tolerates ambiguity, omissions and errors
  in users request

25
Internet AgentsApplications
26
Exploiting Metainformation

• Hotlist title and URL
• Hotlist of 100 items is not the answer!
• Jasper title, URL, keywords, summary, date,
  annotation, ...
• Trade-off go beyond hotlists without copying
  remote information
• Use a richer set of meta-information to index on
  remote information

27
Jasper Agents

• One on each users WWW browser
• Holds a personal profile on each user
• Adapts profile with usage
• Shares information with other users
• Information organiser - interest groups, keyword
  retrieval

28
Storing important information

• Now where did I see ...?
• User asks agent to store interesting information
• Jasper stores a summary keywords locally
• Summary used later to decide whether to retrieve
  remote information
• Keywords used for retrieval

29
Storage

• User requests Jasper agent to store a page
• Agent automatically extracts keywords summary
• User can
• supply an annotation
• post page to a Jasper interest group
• Meta-information stored in Jaspers page store

30
Storage - indexing

• stopping - deletion of common words
• stemming - suffix stripping
• document/term matrix M constructed
• M(i,j) n
• Term (keyword) i occurs n times in document j

31
Storage
Summary Keywords Location (URL) Annotation Interes
t Group User Date
JPS files (meta-info)
Term-1
User Profile
User Profile
Term-2
Term-n
User Profile
32
Sharing Information

• I really must show John...
• When information is stored, agents examines other
  users profiles
• User with relevant interests alerted automatically

33
Sharing Information

• On storing a page, agent checks other users
  profiles
• Profile treated as a query - page scored against
  profile
• coordination level matching score(d) n(keys
  in d)/n(keys in q)
• Agent generates email message to selected users
• URL, annotation and keywords relevant to that
  user are mailed

34
Information Sharing

• Key to future work-styles
• virtual businesses
• distributed teams
• New revenue streams, opportunities from new
  work-styles

35
Agent Learning

• On storage, page matched against users profile
• If no match, agent suggests new keywords
  extracted from information
• Most commonly occurring terms
• User can accept/reject/add
• Users profile evolves over time

36
Retrieval - 1

• via keywords, user and date Show me all the
  pages stored by Tom about VRML this month
• coordination level matching score(d) n(keys
  in d)/n(keys in q)
• contents-addressable information
• information can be relevant in gt1 contexts
• directory/filename structures are ill-equipped

37
Retrieval - 2

• Whats New
• latest pages stored
• those which match profile well
• most recent pages not matching profile
• Interest Groups
• shared lists of links

38
Proactive searching

• Jasper can exploit user profile to search
• Clustering of keywords into related groups
• sim(i,j) 2 ndocs(i,j) / (ndocs(i) ndocs(j))
• Automatic searching - Jasper proactively suggests
  new pages to user

39
Proactive Searching - 2

• Profile internet, information, SMART,
  clustering, agent, intelligent, CORBA, IDL, DCE
• corba dce idl internet .....
• corba 1
• dce 0.62 1
• idl 0.92 0.50 1
• internet 0.03 0.03 0.03 1
• ....
• Complete-link clustering -gt dendogram

40
Proactive Searching - 3
41
Proactive Searching - 4

• dce, corba, idl
• SMART clustering
• internet information
• intelligent agent
• Appropriate thresholds, clustering techniques
• Test on a range of profiles

42
Document Clustering

• Clustering of documents into related groups
  sim(i,j) 2 nterms(i,j) / (nterms(i)
  nterms(j))
• VRML front-end to Jasper store
• emphasis on organistion and display rather than
  search
• idea of scope of collection
• query formulation is not an issue
• document similarities are clearer
• part of a larger initiative - virtual shared
  spaces

43
Summary

• Jasper - an information agent for WWW
• use of meta-information - shared, enhanced
  bookmarks
• information sharing and organisation
• adaptive - user profile learning
• proactive - automatic searching
• WWW - an information sharing (not only serving)
  medium

44
Jasper Agents

• One on each users WWW browser
• Holds a personal profile on each user
• Adapts profile with usage
• Shares information with other users
• Information organiser - interest groups, keyword
  retrieval

45
Jasper Information Agent
Summary Keywords Location (URL) Annotation Interes
t Group
Store retrieve
Share
Profile

• Telecoms
• ATM
• ISDN
• Broadband Services

Search
WWW
46
Internet AgentsKey Challenges

• for static agents the key challenge is keeping
  their indexes up-to-date
• hence future internet agents are likely to be
  mobile
• other challenges similar to those for interface
  and mobile agents

47
Conclusions

• Sharing Information is vital for any organisation
• These lectures attempt to show how this can be
  achieved effectively with agent technology
• We are in the area where agents merge with the
  Semantic Web

48
Sourcehttp//www.firstmonday.org/issues/issue4_9/
odlyzko/index.html
XML Pitfalls