Quality, Relevance and Importance in Information Retrieval with Fuzzy Semantic Networks

Quality, Relevance and Importance in Information Retrieval with Fuzzy Semantic Networks

TMRA 2008 4th International Conference on Topic Maps Research and Applications 15 -17 October 2008 Leipzig Germany

Dino Karabeg, OMS Group, Department of Informatics, University of Oslo

Roy Lachica, Bouvet ASA

Sasa Rudan, HeadWare Solutions



Topic Maps 2008, Oslo

Subject


Roy Lachica

The author of FUZZZY

Knowledge is fuzzy!


Sale in Fisherman’s WorldNorwegian

economy

Fishing

Not all associations are relevant


Jigsaw puzzle idea of knowledge

6


Building Blocks of a Solution

• Accumulate information on usefulness in a value matrix

• Use user point of view or scope to estimate usefulness

• Estimate usefulness by an algorithm


Value Matrix

Criteria

Ways ofevaluating




















12


Agenda

• Goal

• Project Scope

• Problems

• Limitations of Topic Maps

• Information Retrieval

• Proposed Solution

• Partial Implementation on fuzzzy.com


Getting the right information at the right time and place

(Enhancing Information Retrieval systems)

The Ultimate Goal

14


Project Scope

• A model based on: Topic Maps

• For use in: Knowledge based systems

• With a: Social collaborative environment

15


The Obstacles(TMRA-07)

Problems with large scale open collaborative Folktologies:• People use different terms/vocabularies

• Language evolve over time

• People mix different domains and different levels of discourse

• People add errors and noise (overlapping, faulty or imprecise)

• People have different views of what is important. (user-centric relevance)

• People have different views of what things are relevant to each other. (topical relevance)

Topic Map Scopes• The above problems is problematic to solve with TM scopes

Users don't share the same world view. Who decides the scopes?

16


Limitations of Topic Maps

• Associations are not weighted: Flat information, no priority

• No notion of the user and the context

• Problems with representing a nuanced user context with Topic Maps scopes:– Explosion in number of TM constructs/assertions

– High demands on computer processing– User context is something else than the domain and

therefore would be somewhat misplaced in a Topic Map?• Export / fragment size explodes

17


Information Retrieval

• The basic view in IR:– We have a set of resources and we want to retrieval

only the most relevant ones

– This view is to simple?

– What we actually want is to retrieve the most valuable. The ones with the best quality which is relevant and important in our current context

18


Understanding Information

• Relevance, Importance, Quality and Context are overlapping and ill defined

19


QRI

Measure for Information Retrival with Topic Maps:

• Quality. The intrinsic value of an information resource as judged by an individual. (Unreliable or not understandable info is valueless, even if it may otherwise be highly relevant or important)

• Relevance. Relevance is the strength of a relation between two subjects as judged by an individual in a given context. (Different persons with different backgrounds might have different opinions about the appropriateness of relations between concepts)

• Importance. Importance reflects the strength of a relation between a user and a subject in a given context. (Context dependant because the perceived importance of a subject changes over

time as the background and setting of the individual change)

20


Assigning QRI

• Manually– Quality added by rating resources

– Relevance set by rating associations• Topic types for context: e.g. project, event, task, location, group

– Importance assigned by setting topics as important (both individual and

collective)

• Automatic– Quality added when highly ranked users author resources

– Relevance added upon simultanous browsing of topics or when following topic associations

– Importance (low degree) is set when ever a user browse or use a topic

21


Ontology

Social item

-name

Resource proxy-title-description-published date

Tag item

-preferred name-synonym

Actual resource

1 1 0..*

0..*

Group User

-timezone

Related subject

Tag is category of resource content

Subject

Sub class

Context item

-name-timespan

Task Event ProjectLocation

0..*

0..*

Tag is important for user/group

Location also have a timespan. E.g. The city Oslo did not exist 1100 years ago

**

Tagged with** *

****

**

** *

Sub class Sub class

Authorship

*

*

22


Sample socio-semantic contextual network

Friend of

Memberof group

Userbrowsed

User sent

message

Resource is favorite

for group

Task must be

performed in

Headquarter is located in

Event resultedin problem

Task is done

at event

User has

performed task

User has plannedto perform task

User attendedevent

Born in

Context items Tag items

Social items

Relevant for /in/at

XTM isbased on

Is part

of

Topic ofevent

Resoure containinformation about

RDF is based on

supports

Resource is often used by

Is authored by

Probleminvolves

subject

Resource is important for user

Important subjectfor user

User has

browsed subject

User has Used subject

for tagging

User oftenbrowse

User has browsed place

Often browsedtogether

Often browsed

together

Resource

Proxy

Task ispart of

Subject is

importantfor group

Subject space

RDF

XML

XTM

ResourceProxy

Topic Mapssupports

User

User

Task

Location

Event

ResourceProxy

Project

Group

Topic Maps

Sem Web

Resource

Resource

Resource

Strengthened assoc(important for user)

Strengthed assoc(topic 2 topic relevance)

23


QRI Implementation

-Location (PSI ,val ) [0..n]

-Identity (PSI ,val ) [0..n]-Activity (PSI ,val ) [0..n]

-Time (PSI ,val ) [0..n]

Importance Context (user)

Subject is importantfor user

in context of

-Location (PSI ,val) [ 0..n]

-Identity (PSI ,val) [ 0..n]-Activity (PSI ,val ) [0..n]

-Time (PSI ,val ) [0..n]

Relevance Context (user )

Subject

Quality is not context dependant

Resource

Quality (user )

Subject is relevantfor subject

in context ofResource

proxy

SubjectUser SubjectTopic Maps technology

Non Topic Maps technology

24


Context implementation

AssocId userId PSI val

234 567 http://xxx 0.8

234 567 http://xx2 0.3

567 321 http://xx3 0.7

....

subject A subject B

25


QRI in Resource Ranking

Resources

Topical relevance

subjectsubject

subject

subject

User centered relevance

Relevance Context

Importance Context

Entry subjects

subject subject

user

Importance Context

Automatically created

Manually created

26


User-centered Resource Ranking

Important subjectfor user

(0.5)

Is part of(0.8)

Supports(0.7)

Subjectof event

(0.9)

Attends(0.5)

Contextual topic: Event

supports(0.3)

Is important to user(0.4)

ResourceResource

RDFTopic Maps

Semantic Web UserTopic Maps

conference2008

Semantic interoper

ability

27


Ranking in the Context of a Specific Topic

28


Partial Implementation on fuzzzy.com

Fuzzzy.com now supports:

• Association ranking (relevance without context)

• Favorite topics (importance without context)

• Recommend bookmarks (Quality)

• Socio-Semantic search

29


Conclusion

• Advantages of Topic Maps and neural semantic network approach– Intuitive in the user interface

– Semi learning/adaptability

– Less work on part of the user

• Further work– Prototyping, benchmarking

– Constrained Spreading Activation resource calculations

– Context optimalization

– Tuning QRI measures by user

30


Thank you

Quality, Relevance and Importance in Information Retrieval with Fuzzy Semantic Networks

Technology