Using IBM Watson to construct semantic models to extract relevant point-of-care information from Cochrane Reviews

Named Entity Recognition and
Semantic Relation Extraction
from Cochrane Review Documents.

Motivation & Approach
 To make Cochrane contents – evidence - more accessible
 Cross referencing information
 finding relevant passages in documents of 200+ pages
 Supporting discovery & search in Cochrane review documents
 Build a foundation for apps to be used in „point of care“ situations
 Extract an ontology based on information contained in the Cochrane library
 „discover“ Cochrane content relevant to a given patient
 Using semantic models (IBM‘s System T) to extract entities &
relations
 diseases, diagnoses, treatments, interventions, medication, drugs,
symptoms, complications
 „… prolonged treatment with vitamin K antagonists reduces the risk of
recurrent venous thromboembolism …. ”
Page  2
L. Chiticariu, R. Krishnamurthy, Y. Li, F. Reiss, and S. Vaithyanathan, “Domain adaptation of rule-based annotators for named-entity recognition tasks,” in
Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, 2010, pp. 1002–1012.
A. Nagesh, G. Ramakrishnan, L. Chiticariu, R. Krishnamurthy, A. Dharkar, and P. Bhattacharyya, “Towards efficient named-entity rule induction for customizability,”
in Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, 2012, pp.
128–138.

Page  3
Semantic models using System T / AQL
•Extract
candidate
features
•Apply filters
•Post processing
Dictionary
Learning
•Extract basic
features
•Combine
features
•Annotate and
canonical form
Named Entity
Recognition
•Combine
modifiers with
entities
•Combine
Relation Hints
with extended
spans
•Normalize
Relation
Identification

Named Entity Recognition / Build Dictionaries
Page  4

Extract Candidate Features – Context Hints
:treatment <is> <more> effective than :treatment
/(administer|apply)ing/ :treatment
/tak(e|ing)/ :drug
:drug <consumption>
/doses used in/ :disease
/(health )?consequences? of/ :disease
/(medication|therapy|treatment) for/ :disease
Page  5
Dictionary
Learning

Dictionary Learning - Postprocessing
 Shorten by intrinsic hierarchy
 „long-term compression therapy“  „compression therapy“
 “probable ischaemic stroke“  „ischaemic stroke“
(use match statistics as heuristic measures)
 Filter using statistics
 remove „exploded terms“
(few original matches, many dictionary matches)
 remove „weak terms“
(few matches, but many matches for parent)
 Type Deduction
 Type candidates from source extractor modules
 „same list“ (entities mentioned within a list)
 comparation patterns: „compare with“, „in combination with“
Page  6
Dictionary
Learning

Semantic Relation Extraction
Page  7

Building Blocks for Relation Extraction
Named Entities
(„calcium channel blockers“, „parkinson‘s
disease“)
Relation Hints
(A was caused by B; A has positive effects on B …)
 tag with relation type: CAUSE, PREVENT,
INCREASE, …
Modifier Hints
(use of A; risk of A; developing A …)
 tag with modifier type: RISK, USE, INFECTION,
GROWTH, …
Page  8
Relation
Identification

Illustrated cases
 List and Bracket Processing
„Other drugs include carbamazepine and newer antiepileptics
(lamotrigine, topiramate and zonisamide) and the atypical
antipsychotics (clozapine, aripiprazole and ziprasidone)“
 Special Cases (for “is a”)
„atypical antipsychotics (clozapine, aripiprazole and ziprasidone)“
„infections, such as malaria and hookworm”
„selenium, vitamin C and other antioxidants“
 Simple direct relations
@PREVENT :entity /(protect|help)s against/ :entity
@PREVENT :entity <can> <adverb> ? prevent :entity
@CAUSE :entity <is> <adverb>? followed by :entity
@CAUSE :entity <can> <adverb>? result in :entity
Page  9
Relation
Identification

Relation Postprocessing
 Combine consecutive modifiers and Named Entities:
„a reduction in the risk of developing A“
 REDUCE.RISK.GROWTH
 Combine Relation Hints and Extended Entity Spans:
„use of calcium channel blockers was associated with a reduction in the risk of developing
parkinson‘s disease“
 USE A CAUSE REDUCE.RISK.GROWTH B
 Simplify („translate“) to create the final Semantic Relation
 A REDUCE RISK B
 calcium channel blockers REDUCE RISK parkinson‘s
disease
Page  10
Relation
Identification

Page  11
Cochrane Ontology Viewer

Page  12
Text passage supporting a relation

Page  13
Document context of the text passage

Page  14
Cochrane Ontology Viewer – other relations in Chen …

Page  15
Effectiveness of drugs …

Page  16
Evidence for drugs reducing risk of thrombosis

Page  17
What else do we know about „ethynilestradiol? “

Page  18
Relations found in several documents

Observations and insights gained …
 Rule based system adequate for medical reports
 Statistical approaches require larger corpora
 Grammatical parsers alone not sufficiently specific
 Domain specific language aids semantic modelling
 Problems encountered, responses (POS)
 Adjective contamination
 Some antiepileptic drugs are marketed specifically for migraine prophylaxis.
 Delimiting entities and relations
 Drug therapy for migraine falls into two categories.
 Patients were likely to reduce the number of their migraine headaches by 50%.
 Efforts commensurate with the text corpus
 Continuous improvement process inherent in our approach
 Building on top of the existing dictionaries and patterns
Page  19

What‘s next?
 Improve AQL extraction results
 Improve entity normalization and types
(eg. make better use of entity components: „endothelin receptor
antagonist“)
 Identify most relevant relations
 Extraction of Structured Context
 Use ontology for point of care situations
 Introduce deep learning technology
 User interface (mobile systems of engagement) for „point of care“
situations
 Combine with patient data to guide the discovery process in
Cochrane reviews
Page  20

Using IBM Watson to construct semantic models to extract relevant point-of-care information from Cochrane Reviews

More Related Content

Viewers also liked (13)

Similar to Using IBM Watson to construct semantic models to extract relevant point-of-care information from Cochrane Reviews (20)

More from Paradigma Consulting (20)

Recently uploaded (20)

Using IBM Watson to construct semantic models to extract relevant point-of-care information from Cochrane Reviews