Knowledge Representation COMP34512 Sebastian Brandt - - PowerPoint PPT Presentation

Knowledge Representation COMP34512

Sebastian Brandt

brandt@cs.manchester.ac.uk http://cs.man.ac.uk/~sbrandt http://studentnet.cs.manchester.ac.uk/syllabus/? code=COMP34512&year=2013 Slides: Bijan Parsia, Bijan Parsia (bijan.parsia@manchester.ac.uk)

Tuesday, 28 January 14

Knowledge Representation

• “Knowledge”

– Comes in different forms

• Knowing that vs. knowing how (or “know-how)
• “Propositional” vs. “procedural”

– Classic philosophical definition: Justified True Belief

• Generally we use a looser sense,

– e.g., “believed to be true” or “asserted”

• Representation

– Anything with “aboutness”

• I.e., a re-presentation has a relation to some other “thing”
• The representation proxies for the thing
• Common case: a symbol (or symbolic structure)
• Knowledge + Representation?

– (“KR” or sometimes “KRep) – Isn’t all representation representation “of knowledge”?

2

Tuesday, 28 January 14

Computational KR

• We’re concerned with

– representing propositional knowledge – symbolically (mostly) – in a way amenable to programatic manipulation

• in particular, amenable to automated reasoning
• This introduces constraints

– Know that + symbols suggests KR formalisms

• I.e., languages designed to support building KRs

– Manipulability suggests

• precise semantics
• computationally reasonable processes
• Our starting paradigm

– Written human communication

• And supporting mechanisms

– I.e., Modelling

3

Tuesday, 28 January 14

Applications

• NLP

– As support or target

• Data integration
• Expert systems

– Classically diagnosis

• Intelligent tutoring systems
• Games
• Cognitive science/Psychology
• Planning and decision support
• Personal assistants
• Data acquisition
• Clinical support systems
• Etc.

4

Tuesday, 28 January 14

COMP34512

• We’ll explore

– the conceptual foundations – various formalisms

• with a focus on design considerations

– modelling techniques – reasoning – applications

• and benefits of using KR
• Core formalism family

– Decidable fragments of FOL – Specifically “description logics”

• Other topics

– Psychology modelling – The Semantic Web & Linked Data

5

Tuesday, 28 January 14

Anatomy of the Course

• Bi-weekly lectures (mostly)

– Tuesday at 12:00 (but you, here, know this!) – Thursday at 12:00

• Associated/recommended reading will be given

– Not all of the relevant information will be contained in lecture materials and notes – Sometimes there is anticipatory reading

• Ask questions and promote discussions in lectures

– I’m quite happy to have (fruitful) digressions – When in doubt, speak

• There will be guest lectures

6

Tuesday, 28 January 14

Assessment

• Exam and coursework
• Exam: 80%

– Online – Consists of a combination of objective and freetext questions

• Coursework/Labs: 20%

– Short essays – Modelling exercises – VLE for distribution and return

• I try to avoid conflicts with other deadlines

– esp. project deadlines – Let me know if things crop up

• before my deadlines!
• In general, deadlines are strict

– Mitigating circumstances @ SSO is the way to go

7

Tuesday, 28 January 14

Resources

• Course web page

– http://studentnet.cs.manchester.ac.uk/syllabus/ ?code=COMP34512&year=2013

• University’s Blackboard system

– Discussion and collection of relevant materials – Discussion Forum

• I monitor closely
• Twitter for course updates and comments

– Hash tags: #uon #comp34512 – Users: @bparsia

8

Tuesday, 28 January 14

A simple analogical example

Tuesday, 28 January 14

A Problem

• Something we know

– What a Google Search Results page is – How to recognise it

• How to capture this knowing?
• First, what’s the content

– (Simplified example!) – Google's pages contain “Google” – Search result pages expand the number of “o”s in “Google”

• Second, how to express it

– And what do we get for different sorts of expression?

10

Tuesday, 28 January 14

In Natural Language

• We already did that!

– “Google's pages contain “Google” and Search result pages expand the number of “o”s in “Google””

• But that’s not very precise or clear

– We can refine it

• A page that has the extended spelling of “Google”
• A page that contains “Gooogle”
• A page that contains a substring “Gooo”
• A page that has a substring “Goo”

– followed by one or more “o”s, – then by “gle”

• Way more precise but

– not directly actionable – we have a spec, not a recogniser

• Precise specs are good things, of course

11

Tuesday, 28 January 14

As Computer Code

• Executable (mostly)
• Precise
• But.....

– Not very clear – Lots of irrelevant details – Portability issues

12

Tuesday, 28 January 14

A Formalism

• We could design a language for strings

– Basic characters: a-zA-Z – * for zero or more

• This allows us to represent a class of strings, i.e.:

– Goooo*gle

• To exploit the representation, we use “search”

– search("Goooo*gle", "Gogle") ⇒ false – search("Goooo*gle", "Google") ⇒ false – search("Goooo*gle", "Gooogle") ⇒ true – search("Goooo*gle", "Goooooogle") ⇒ true

• We don’t have to care how “search” is implemented

13

Tuesday, 28 January 14

Extending the Formalism

• We could add expressivity

– More characters: 0-9! – | for alternation

• We might design other services

– equiv("(Yah|G)oooo*(gle|!)","Yahoooo*!|Goooo*gle") – contained_in("Yahooo*!","(Yah|G)oooo*(gle|!)")

• We can evaluate our formalism

– Can we say what we want to say? – Can we do useful things with what we say? – Is it easy to say what we want to say

• or understand what we’ve already said?

14

Tuesday, 28 January 14

Next Class

• We'll start on the knowledge acquisition problem

– How to conceptualise a domain

• I.e., the refinement step
• In class lab

– Please attend and on time!

15

Tuesday, 28 January 14