Indexation automatique de documents en santé : évaluation et analyse de sources d’erreurs

Chebil, Wiem; Soualmia, Lina Fatima; Dahamna, Badisse; Darmoni, Stéfan Jacques

doi:10.1016/j.irbm.2012.10.002

Cited by 5 publications

(7 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…These two IR models [8,9] do not use semantic resources, which leads to less precision. The possibility and necessity measures are used to map the query to the documents [2]. For document ranking, Ref.…”

Section: Related Workmentioning

confidence: 99%

“…Moreover, IMSIR uses a multinomial naive Bayes classifier (MNBC) to extract concepts. The MNBC allows the IMSIR to enrich the index with new concepts whose terms do not occur in the documents [2]. For example, the concept "Bronchodilator Agents" belongs to the index of the document having the PMID = 11115306, although this concept does not occur in the document.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Improving Semantic Information Retrieval Using Multinomial Naive Bayes Classifier and Bayesian Networks

et al. 2023

View full text Add to dashboard Cite

This research proposes a new approach to improve information retrieval systems based on a multinomial naive Bayes classifier (MNBC), Bayesian networks (BNs), and a multi-terminology which includes MeSH thesaurus (Medical Subject Headings) and SNOMED CT (Systematized Nomenclature of Medicine of Clinical Terms). Our approach, which is entitled improving semantic information retrieval (IMSIR), extracts and disambiguates concepts and retrieves documents. Relevant concepts of ambiguous terms were selected using probability measures and biomedical terminologies. Concepts are also extracted using an MNBC. The UMLS (Unified Medical Language System) thesaurus was then used to filter and rank concepts. Finally, we exploited a Bayesian network to match documents and queries using a conceptual representation. Our main contribution in this paper is to combine a supervised method (MNBC) and an unsupervised method (BN) to extract concepts from documents and queries. We also propose filtering the extracted concepts in order to keep relevant ones. Experiments of IMSIR using the two corpora, the OHSUMED corpus and the Clinical Trial (CT) corpus, were interesting because their results outperformed those of the baseline: the P@50 improvement rate was +36.5% over the baseline when the CT corpus was used.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Improving Semantic Information Retrieval Using Multinomial Naive Bayes Classifier and Bayesian Networks

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Le but de cette étape de filtrage est de ne garder que les concepts pertinents parmi ceux ayant un sous-ensemble des mots de leur terme représentatif qui n'existe pas dans le document. En effet, nous avons classé la non-extraction de ces concepts pertinents comme une catégorie d'erreurs d'indexation dans une étude antérieure (Chebil et al, 2012). Cette étape consiste à diviser l'ensemble de concepts extraits dans l'étape précédente en deux ensembles de concepts.…”

Section: éTape 3 Filtrageunclassified

Extraction possibiliste de concepts MeSH à partir de documents biomédicaux

Chebil¹,

Soualmia²,

Omri³

et al. 2014

Revue d'intelligence artificielle

View full text Add to dashboard Cite

RESUME. Nous proposons dans cet article une nouvelle approche d'indexation de documents biomédicaux basée sur les réseaux possibilistes permettant de les apparier partiellement aux termes du thésaurus MeSH (Medical Subject Headings). La principale contribution de notre approche est le traitement de l'imprécision et de l'incertitude liée à la tâche d'indexation à travers l'utilisation de la théorie des possibilités. En effet, nous proposons d'améliorer l'estimation de la pertinence d'un document étant donné un concept en utilisant deux mesures : la possibilité et la nécessité. La possibilité estime le degré de rejet d'un document non pertinent étant donné un concept. La nécessité de la pertinence d'un document évalue dans quelle mesure un document est pertinent pour le concept. De plus, notre approche permet de réduire les limites de l'appariement partiel qui génère de l'information inutile, bien que ce type d'appariement permette de trouver dans le document d'autres variantes du vocabulaire contrôlé. Pour ce faire, nous proposons de filtrer l'index en utilisant les connaissances fournies par l'UMLS (Unified Medical Langage System). Filtrer l'index permet de ne garder que les concepts pertinents parmi ceux ayant un sous-ensemble de leurs mots dans le document. Les expérimentations réalisées sur les deux corpus OSHUMED et CISMeF ont montré des résultats encourageants. ABSTRACT. We propose in this paper a new approach for indexing biomedical documents based on the possibilistic network, which carries out a partial matching between documents and the MeSH thesaurus (Medical Subject Headings) terms. The main contribution of our approach is to deal with the imprecision and the uncertainty of the indexing task by using the possibility theory. In fact, we propose to enhance the estimation of a document relevance given a concept by using the two measures of possibility and necessity instead of only one measure used by common approaches. The possibility measure estimates the degree of rejection of an irrelevant document given a concept. The necessity of the relevance of a document estimates what extent a document is relevant for a given concept. Our contribution also consists in reducing the limitation of the partial matching that generates irrelevant information although it allows finding in the document other variants of terms than those in the dictionaries. In fact, we propose to filter the index using the knowledge provided by the Unified Medical Language System (UMLS). The filtering allows keeping relevant concepts among those having a subset of their words terms in the document. The experiments carried out at the different steps of our approach and on different corpora showed very encouraging results. MOTS-CLES : indexation de documents biomédicaux, réseaux possibilistes, appariement partiel, vocabulaires contrôlés.

show abstract

“…In addition, more recently there was AgNIC (Agriculture Network Information Collaborative), which bases automatic indexing on the Thesaurus, as Luxid does. • CISMeF (Chebil et al, 2012). The Catalogue et Index des Sites Médicaux de langue Française, a system implemented for the automatic indexing of medical information resources.…”

Section: Introductionmentioning

confidence: 99%

Indización automática de artículos científicos sobre Biblioteconomía y Documentación con SISA, KEA y MAUI

Gil-Leiva

Ortuño²,

Corrêa³

2022

revespdoccient

View full text Add to dashboard Cite

Este artículo evalúa los sistemas de indización automática SISA (Automatic Indexing System), KEA (Keyphrase Extraction Algorithm) y MAUI (Multi-Purpose Automatic Topic Indexing) para averiguar cómo funcionan en relación con la indización realzada por especialistas. El algoritmo de SISA se basa en reglas sobre la posición de los términos en los diferentes componentes estructurales del documento, mientras que los algoritmos de KEA y MAUI se basan en el aprendizaje automático y las frecuencia estadística de los términos. Para la evaluación se utilizó una colección documental de 230 artículos científicos de la Revista Española de Documentación Científica, publicada por el Consejo Superior de Investigaciones Científicas (CSIC), de los cuales 30 se utilizaron para tareas formativas y no formaban parte del conjunto de pruebas de evaluación. Los artículos fueron escritos en español e indizados por indizadores humanos utilizando un vocabulario controlado en la base de datos InDICES, también perteneciente al CSIC. La indización humana de estos documentos constituye la referencia contra la cual se evalúa el resultado de los sistemas de indización automáticos, comparando conjuntos de términos usando métricas de evaluación de precisión, recuperación, medida F y consistencia. Los resultados muestran que el sistema SISA funciona mejor, seguido de KEA y MAUI.

show abstract

Indexation automatique de documents en santé : évaluation et analyse de sources d’erreurs

Cited by 5 publications

References 18 publications

Improving Semantic Information Retrieval Using Multinomial Naive Bayes Classifier and Bayesian Networks

Improving Semantic Information Retrieval Using Multinomial Naive Bayes Classifier and Bayesian Networks

Extraction possibiliste de concepts MeSH à partir de documents biomédicaux

Indización automática de artículos científicos sobre Biblioteconomía y Documentación con SISA, KEA y MAUI

Contact Info

Product

Resources

About