MELISA

1
MELISA

• An ontology-based agent for information retrieval
  in medicine

Jose Maria Abásolo Mario Gómez Institut
dInvestigaciò en Intel.ligència Artificial
(IIIA) Spanish Scientific Research Council
(CSIC)
2
Index

• Motivation
• Overview
• MELISA process
• Query Generation
• Query Evaluation, Filter Combination
• Results
• Conclusions
• Future Work

3
Motivation

• Nowadays Internet gives us a great quantity of
  information
• Most users find difficult to formulate
  well-designed queries for retrieval purposes
• Usually a user makes a first query and then he
  has to reformulate the query (one or more times)
  to get useful information
• This project try to solve this problem within a
  professional domain (biomedical literature)

4
Overview
5
Medical Ontology
6
GUIDELINES is-an-instance-of EVIDENCE_INTEGRATION
Name Guidelines MeSH_Terms Guidelines,
Practice Guidelines, Clinical Protocol
Publication_Type guideline, practice guideline
Related_MeSH_Terms Guideline
Adherence
7
Query Model
Queries valid for some data source
Very abstract, is given by the user
Link the consultation to the ontology
8
Generation of queries
9
(No Transcript)
10
Pneumonia Ofloxacin
Decomposition Level 1
Good Evidence
Therapy
EBM
Cost Analysis
Guidelines
Decomposition Level 2
..
Specific Query1
Specific Query2
Specific Query n
Specific Query3
SQ1 pneumonia ofloxacin AND guidelines MAJR
SQ2 pneumonia ofloxacin AND guidelines
MHNOEXP SQ3 pneumonia ofloxacin AND
guidelines MH
11
Query evaluation combination

• Scoring documents inside a Conceptual Query
• Combine documents from different conceptual
  queries

12
Scoring documents inside a Conceptual Query
LIST UID
SPECIFIC QUERY
Weighted Sum
LIST UID
SPECIFIC QUERY
CONCEPTUAL QUERY
LIST UID
LIST SCORED UID
SPECIFIC QUERY
LIST UID
SPECIFIC QUERY
LIST UID
SPECIFIC QUERY
13
Combine documents from different Conceptual
Queries
Categories To Combine
List of Documents
14
Combine documents from different Conceptual
Queries (II)
LIST SCORED UID
CONCEPTUAL QUERY
Aggregation Function
LIST SCORED UID
CONCEPTUAL QUERY
LIST SCORED UID
CONCEPTUAL QUERY
LIST OF DOCUMENTS
LIST SCORED UID
CONCEPTUAL QUERY
LIST SCORED UID
CONCEPTUAL QUERY
15
Results

• Comparison between MELISA and a human user
  working with PubMed
• 5 queries (evaluating best 40 documents for any
  query)
• For example
• Human user query
• Osteoporosis AND Women AND (Therapy OR Guideline
  OR Cost)
• MELISA
• Keywords Osteoporosis, Women
• Selected categories Therapy, Guideline, Cost
  analysis

16
Results (II)
17
Conclusions

• The system is able to integrate a big amount of
  information and show the results in a dynamic way
• The use of the ontology has two main benefits
• Helps user to make a consultation
• Allow to use synonymous and related terms
• Our architecture seems to be a good approach to
  solve the problem of domain and source
  independence, but it needs to be improved
• A great problem is the combination of results
  from different categories
• The first empirical test shows that the system
  improves the traditional retrieve using PubMed

18
Future work

• To develop user profiles
• To work with multiple information sources
• To study and compare different evaluation
  functions
• To study more complex criteria to reformulate the
  specific queries
• To develop algorithms for learning the weight
  coefficients
• To apply the system in other domains
• To study other query (reformulation) operators
  (generalization, specification, source selection)