1. Kenneth Baclawski et. al. PSB 2000 2002/11/7 Sa-Im Shin
Knowledge Representation and Indexing using the Unified Medical Language System
Kenneth Baclawski et. al.
PSB 2000
2002/11/7
Sa-Im Shin

2. Contents Introduction Related Work
Using the UMLS to Express Document Semantics
Constructing Knowledge Representations
Semantic Based User Interfaces
Usability Survey
Survey Results
Conclusion
After introduction and related work, I’ll explain the method of using ~, Constructing ~ and semantic ~.
Then we’ll talk about usability ~ and survey ~, finally make conclusion.

3. Introduction Document Understanding Syntactic understanding
Lexical scanning, Morphological Analysis, Parsing
Semantic Understanding : Ontology
Framework for understanding & representing knowledge
Meaning of the vocabulary terms <- Vocabulary in the community
Unified Medical Language System (UMLS)
Rich & well structured ontology in biomedical domain
Need for adaptation for application
There are tow kind of document ~.
One is the syntactic ~ like lexical ~.
Our focused understanding is semantic ~ especially ontology.
Ontology is the framework ~.
And meaning ~ is represented by vocabulary ~.
Unified ~ is the one of the famous ontology in bioinformatics.
UMLS is rich ~, but it need some processing for adaptation ~.

4. Related Work Ontology-based techniques
Representing, indexing & retrieving documents
First stage (1992 ~): Relatively small ontology
World Wide Web (1996~ )
General large ontology
OntoSeek, InQuizit
Ontology for biological knowledge
Hafner, Fridman, Schulze-Kremer
UMLS
Larger than OntoSeek
Knowledge representation by text processing
Using typical NLP techniques
Since 1990, there have been many ontology-based effort for representing, ~.
At first, relatively small ontology is constructed.
With the spread of world ~, general and large ~ appears like OntoSeek and InQuizit.
Hafner, ~ are ontology ~.
And UMLS is larger ~ using ~

5. Using the UMLS to Express Document Semantics
Ontology components for knowledge representation
Semantic categories (types) : 130
Semantic relationships : 50
Categorical links : 7000
Semantic concepts : 475,000
Concept maps : Mapping from source data to semantic concept (1,000,000)
Categorization : Semantic concept to semantic category (600,000)
Conceptual links : Concepts & Relationship (372,808)
Explanation
Knowledge Representation Rules
Ontology ~ are like these.
The numbers in this list are the number in the UMLS.

6. Constructing Knowledge Representations
System interface
Input : Document
Output : visual representation
Background knowledge
Domain-independent knowledge
Syntactic knowledge, POS & grammar rules
Domain-dependent knowledge
Semantic knowledge, Terminology of the domain
This paper classifies background ~ into domain-~ and domain-~.
Domain-~ contains syntactic ~, and semantic ~ are domain-.

7. Constructing Knowledge Representations
Lexeme -> conceptual units (tokens)
Original source data -> semantic concept of ontology
Extracting elementary units (lexeme)
Extracting protein & chemical names
Parse tree & anaphoric references -> knowledge representation
Knowledge representation rules : Condition & action
Fragmenting & indexing
Distributed high-performance indexing engine
Sequence of tokens -> parse tree
Anaphora construction : Pronoun -> Anaphoric references
Reducing ambiguity by ontology
This figure show the whole processing for constructing ~.
First scanning module, we ~, used domain independent ~ is word ~ and domain-dependent ~ is the domain-~.
Then throughout tokenization, we ~. We use ~ in domain-independent ~ and ~ in domain-dependent ~.
In parsing ~, we ~ with domain-independent grammar ~ and domain-dependent relationship ~, and the result of this module are parse ~.
With these parse ~ we represent knowledge with knowledge ~ containing various condition ~.
And for fragmenting ~, we use distributed ~.

8. Semantic Based User Interfaces
Developing user interface
Insight into the queries & retrieved data
Keynets
Diagrammatic knowledge representation
User interface is developed by diagrammatic ~ keynets to help understand queries ~.

9. Usability Survey Line Box question, description of
a pharmacological product
Medline keywords in the text
appearing in the UMLS
Corresponding biomedical
keynet representation
of biomedical text
Line
Relationship between two concepts
Label : relationship type (property)
Dot : Source of relationship (statement)
cf) object / predicate
This figure shows the shape of user survey.
Diagram in the center is keynet representation.
Box represents ~ with word ~.
Line shows the relationship ~.
Label express relationship ~ like “part of ~”, and dot mark source ~.
And especially in the case of “part of”, we divide boxes into object ~.
Survey contains three example.
Biomedical text is a question, ~, biomedical keywords are medline ~, and previous biomedical keynet.
Box
Representing semantic concept
Word or short phrase in UMLS

10. Usability Survey Feedback Semantic differential Feedback for keynet
Interpreted nine questions
Gauging the subject’s overall reaction
0 to 9
Feedback for keynet
Using semantic differential
Presentation of keynet
Comparison of three examples
Also we feedback our results using semantic ~.
Semantic ~ is interpreted ~ to gauge ~ and the scale is 0 ~.
Feedback evaluate the presentation ~ by comparison ~.

11. Survey Results Ease of Understanding Limit complexity
High understandability
Limit complexity
More detailed views when user’s request
Keyword vs. Keynet
Keynet is more useful than keyword in some cases
These survey results help easy understanding and limit complexity for high ~.
More ~ are provided when ~.
And keynet ~.

12. Conclusion Role & requirements of ontology Evaluation
Constructing knowledge representation
NLP tools for Using UMLS ontology
Evaluation
Usefulness of representing biomedical knowledge
Favorable results
As a conclusion,
We construct knoledge ~ and NLP ~.
and we evaluste the usefulness ~ then got favorable ~.

13. Future Works Short term goal Long term goal Reflecting Feedback
Development of new class of semantic-based search engine
New usability survey
Long term goal
Dynamic interaction with user
Expanding subject areas
Short term future ~ are reflecting ~ to ~ and new ~.
Long term goal is implementing dynamic ~ and expanding ~.

14. Discussion Issue
I think this paper lacks of evaluation part. In fact, evaluation for ontology and representation is difficult problem.
Do you have good idea about evaluation methods for these topics ?
What do you think about important evaluated faces in these topics ?
In this paper, system provide user feedback for developing system.
This feedback results contains user’s evaluation and grade of various aspects about the system.
And I think usefulness of the expression in other application system is also very important factor.
So, we can apply this result into input of other application system, and can compare the performance with other representation or without this result.