32
Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S. National Library of Medicine Artificial General Intelligence Research Institute Workshop

Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Embed Size (px)

Citation preview

Page 1: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Knowledge-Based Semantic Interpretationfor Summarizing Biomedical Text

Thomas C. Rindflesch, Ph.D.Marcelo Fiszman, M.D., Ph.D.

Halil Kilicoglu, M.S.

National Library of Medicine

Artificial General Intelligence Research Institute Workshop

Page 2: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Overview

Symbol grounding Meaning consists of the manipulation of an internal

system of relationships among concepts (Rapaport 1995)

Illustrate the viability of this approach Semantic interpretation for biomedical research literature

Suggest that the system adumbrates intelligence Provides the basis for reasoning about medical topics

Page 3: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Unified Medical Language System (UMLS)

Developed at the National Library of Medicine Compilation of more than 100 terminologies in the

biomedical domain Two domain knowledge components

Metathesaurus: concepts Semantic Network: relationships

Constitutes the “meaning” of medicine Incomplete Inconsistent Useful

Page 4: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Metathesaurus

More than 1,000,000 concepts in biomedicine Disorders Organisms Anatomy, physiologic functions Drugs, procedures

Synonyms Hierarchical structure Each concept assigned semantic types (or

categories)

Page 5: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Metathesaurus Concept

Drug Therapy, Combination; Combination Chemotherapy;Polychemotherapy

Therapeutic or Preventive Procedure

•Analytical, Diagnostic and Therapeutic Techniques and Equipment

•Therapeutics•Drug Therapy

Page 6: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Metathesaurus Concept

Mycoplasma pneumonia; Eatons agent pneumonia;Endemic pneumonia; et al.

Disease or Syndrome

•Respiratory Tract Diseases • Lung Diseases

•Pneumonia•Pneumonia, Bacterial

Page 7: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Network

134 semantic types Disease or Syndrome Therapeutic or Preventive Procedure Pharmacologic Substance Body Part, Organ, or Organ Component

In two hierarchies: Entity, Event

54 Relationships between semantic types

Bacterium - CAUSES - Pathologic Function

Pathologic Function - PROCESS_OF - Organism

Page 8: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

affects

functionally_related_to

brings_about

physicallyspatially

temporallyconceptually

associated_withSemantic Network Predicates

occurs_in

Page 9: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

TREATS

affects

functionally_related_to

brings_about

physicallyspatially

temporallyconceptually

associated_withSemantic Network Predicates

CO-OCCURS_WITH

PREVENTS

OCCURS_IN

CAUSES

LOCATION_OF

Page 10: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

affects

functionally_related_to

brings_about

physicallyspatially

temporallyconceptually

associated_withSemantic Network Predication

occurs_in

Occupational Activity

Health Care Activity

Therapeutic or Preventive Procedure

Disease or Syndrome

Biologic Function

Pathologic Function

treats

Page 11: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Interpretation: SemRep

Exploit the UMLS for processing medical text Interpret (some of) the meaning asserted in

language Map words in language to concepts

Metathesaurus

Use syntactic structure to identify relationships between concepts Semantic Network

Page 12: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

SemRep Output

Mycoplasma pneumonia is an infection of the lung caused by Mycoplasma pneumoniae.

Mycoplasma Pneumonia ISA Infection Lung LOCATION_OF InfectionLung LOCATION_OF Mycoplasma PneumoniaMycoplasma pneumoniae CAUSES Infection Mycoplasma pneumoniae CAUSES Mycoplasma Pneumonia

Page 13: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Related Research in Biomedicine

BioMedLEE, GENIES Semantic grammar

AQUA Definite clause grammar

MPLUS Chart parser

MEDSYNDIKATE Dependency grammar

[Friedman, et al.]

[Haug, et al.]

[Johnson, Campbell]

[Hahn, et al.]

Page 14: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Lexical semantics Contribution of words to interpretation

Meaning-text theory Network of semantic predications Syntax rules are interpretive devices

Ontological semantics Applied interpretation Ontology is the main metalanguage of meaning

Semantics Framework

[Mel’cuk]

[Nirenburg & Raskin]

[Cruse; Pustejovsky]

Page 15: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

SPECIALISTLexicon

MetaMap ParserMetathesaurus

SemRep: System Overview

SemanticNetwork

ConstructRelation

MedicalText

MedPostTagger

LexicalLook-up

ResolveAmbiguity

SemanticPredication

Page 16: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Input

The aim of this study was the characterization of the specific effects of alprazolam versus imipramine in the treatment of panic disorder with agoraphobia and the delineation of dose-response and possible plasma level-response relationships.

Page 17: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

SPECIALISTLexicon

Parser

Syntactic Processing

TextMedPostTagger

LexicalLook-up

ResolveAmbiguity

Page 18: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Syntactic Processing

The aim of this study was the characterization of the specific effects

NP[of alprazolam] [versus] NP[imipramine]

NP[in the treatment]Nominalization

NP[of panic disorder] NP[with Agoraphobia] and the delineation of dose-response and possible plasma level-response relationships.

Page 19: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

MetaMap: Metathesaurus Concepts

SPECIALISTLexicon

MetaMap ParserMetathesaurus

TextMedPostTagger

LexicalLook-up

ResolveAmbiguity

Page 20: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

MetaMap: Metathesaurus Concepts

The aim of this study was the characterization of the specific effects

NP[of Alprazolam] [versus] NP[Imipramine]

NP[in treatment]Nominalization

NP[of Panic Disorder] NP[with Agoraphobia] and the delineation of dose-response and possible plasma level-response relationships.

Page 21: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Types

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

Pharmacologic Substance

Disease or Syndrome

Page 22: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Construct Predication

MetaMap ParserMetathesaurus

SemanticNetwork

ConstructRelation

MedicalText

MedPostTagger

LexicalLook-up

ResolveAmbiguity

SemanticPredication

SPECIALISTLexicon

Page 23: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Interpretation

Indicator rules Establish a link between

Words in text Predicates in the Semantic Network

Argument identification rules Syntactic constraints

Interpretation of semantic predications UMLS Semantic Network

Page 24: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Indicator Rules

inpreposition TREATSHemofiltration in digoxin overdose

inpreposition HAS_LOCATION

Severe infections in both feet

Establish a correspondence between a syntactic item and a Semantic Network predicate

ItemStructure Semantic Network

treatment TREATS

Drugs for the treatment of schizophrenia

nominalization

Page 25: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Types

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

Pharmacologic SubstanceDisease or Syndrome

Page 26: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Apply Indicator Rule

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

TREATS

Page 27: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Argument Constraints

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

TREATS

Page 28: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Semantic Network Predication

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

medd-TREATS-dsyn

phsu-TREATS-dsyn

topp-TREATS-dsyn

topp-TREATS-inpo

Page 29: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Match Semantic Types

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[phsu] NP[in treatment]Nominalization

NP[of dsyn] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

medd-TREATS-dsyn

phsu-TREATS-dsyn

topp-TREATS-dsyn

topp-TREATS-inpo

Page 30: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Substitute Concepts

The aim of this study was the characterization of the specific effects

NP[of phsu] [versus] NP[Alprazolam] NP[in treatment]Nominalization

NP[of Panic Disorder] NP[with dsyn] and the delineation of dose-response and possible plasma level response relationships.

Alprazolam-TREATS-Panic Disorder

Page 31: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Manipulate Predications

Abstraction summarization on a given topic Treatment of disease

Apply to predications from multiple documents Devise summarization rules

Relevance: “Stick to the point” Predications adhere to a schema for treatment of disease

Novelty: “Don’t tell me what I already know” Eliminate arguments high in the UMLS hierarchy

Salience: “Give me the main points” Eliminate low frequency predications

[Hahn]

Page 32: Knowledge-Based Semantic Interpretation for Summarizing Biomedical Text Thomas C. Rindflesch, Ph.D. Marcelo Fiszman, M.D., Ph.D. Halil Kilicoglu, M.S

Summary Results

Search Medline Limit to previous year: 294 citations

Summarize retrieved documents Provide an informative overview

Further reasoning on the summarized predications is feasible