How do users formulate their queries? A morpho-syntactic analysis - - PowerPoint PPT Presentation
11th European Conference of Medical and Health Libraries How do users formulate their queries? A morpho-syntactic analysis Nicolas Ariste Fairon Life sciences library, University of Liege, 4000 Lige, Belgium <nicolas.fairon@ulg.ac.be>
Nicolas Ariste Fairon Life sciences library, University of Liege, 4000 Liège, Belgium <nicolas.fairon@ulg.ac.be>
24th of June, Nicolas Fairon
2
Natural Language Processing Automatic extraction of concepts
3
Despite the efforts, many users remain unable to perform
Introduction – Material & Methods – Results - Conclusions
Bad query formulation Bad knowledge of MeSH terms Not enough practice Problems with boolean operator
4
Medline interfaces, with interesting features:
Query expansion
Searching MeSH and keywords Automatic explosion...
Permuted index MeSH translations Elementary tools for natural language searching
Introduction – Material & Methods – Results - Conclusions
5
Introduction – Material & Methods – Results - Conclusions Medline interfaces Natural language
Precision Recall Controlled language
83.7% 100%
Torticollis [MeSH]
40.0% 90.0%
Torticollis/cn [MeSH]
44.1%
Smoking/ae [MeSH]
complexity Efficiency
6
Introduction – Material & Methods – Results - Conclusions
7
Query submitted by user Corrected Semantically tagged CORPUS All queries Analysis Descriptive Concepts extraction Dictionary Local grammar Hybrid Approaches
Introduction – Material & Methods – Results - Conclusions
Manual Automatic
8
Query submitted by user Corrected Semantically tagged
Introduction – Material & Methods – Results - Conclusions
9
To append semantic flags to useful concepts To identify and keep track of every concept To evaluate the efficiency of our application
Introduction – Material & Methods – Results - Conclusions
Query submitted by user Corrected Semantically tagged
10
Query submitted by user Corrected Semantically tagged CORPUS All queries
Introduction – Material & Methods – Results - Conclusions
A web application to store for each query
Raw, corrected, and tagged versions Medline search history done by a scientific librarian
195 queries formulated by 68 different users 6 985 words
11
Analysis Descriptive
Introduction – Material & Methods – Results - Conclusions
Concepts extraction Dictionary Local grammar Hybrid UNITEX Dictionnaries French MeSH Hand-made Local grammars
12
Introduction – Material & Methods – Results - Conclusions
List B (reference) List A COMPARISON Recall
Precision
Concepts extraction Queries Concepts CORPUS u n t a g g e d tagged VS
13
464 concepts have been identified
Introduction – Material & Methods – Results - Conclusions
14
Applying MeSH dictionary to queries in order to
Introduction – Material & Methods – Results - Conclusions
MeSH terms Subheadings Keywords
10 20 30 40 50 60 70 80 90 100
Recall Precision
%
15
Use recognition patterns relying on
Introduction – Material & Methods – Results - Conclusions
MeSH terms Subheadings Keywords
10 20 30 40 50 60 70 80 90 100
Recall Precision
%
16
Using local grammars combined with dictionaries
Introduction – Material & Methods – Results - Conclusions MeSH terms Subheadings Keywords
10 20 30 40 50 60 70 80 90 100 Recall Precision %
17
Creating a new interface based on natural language
Concept mapping Concepts combination
Hybrid approach shows best results
Dictionaries Local grammar
Dictionaries'quality influes on performance
Introduction – Material & Methods – Results - Conclusions
18
Disambiguiation of fuzzy MeSH concepts Combination of the concepts with adequate booleans
Made the tool available to users as a web application
Introduction – Material & Methods – Results - Conclusions
19
Open source tools used for the work and the presentation :