Towards the development of a conceptual distance metric for the UMLS

Abstract

The objective of this work is to investigate the feasibility of conceptual similarity metrics in the framework of the Unified Medical Language System (UMLS). We have investigated an approach based on the minimum number of parent links between concepts, and evaluated its performance relative to human expert estimates on three sets of concepts for three terminologies within the UMLS (i.e., MeSH, ICD9CM, and SNOMED). The resulting quantitative metric enables computer-based applications that use decision thresholds and approximate matching criteria. The proposed conceptual matching supports problem solving and inferencing (using high-level, generic concepts) based on readily available data (typically represented as low-level, specific concepts). Through the identification of semantically similar concepts, conceptual matching also enables reasoning in the absence of exact, or even approximate, lexical matching. Finally, conceptual matching is relevant for terminology development and maintenance, machine learning research, decision support system development, and data mining research in biomedical informatics and other fields.

A Cluster-Based Approach for Semantic Similarity in the Biomedical Domain

“…The least common subsumer (LCS) of two concept nodes determines the common specificity of two concept nodes (e.g. LCS(a 2 ,a 6 )=a 1 & LCS(a 1 ,b 1 )=r), therefore we use LCS for computing common specificity of two concept nodes. Furthermore, local density such as link strength/weight also affects the similarity.…”

“…Let us consider, for example, a fragment of ontology showing two clusters as in Figure 1. We define the specificity of a concept c in cluster C as follow: depthC depth(c) spec(c) (1) where depthC is the depth of cluster C, and spec(c) [0,1]. We notice that spec(c) = 1 when the concept c is a leaf node in the cluster C. Then, in Figure 1, the specificity of a 3 and b 3 , is calculated as follow: spec(a 3 )=3/4= 0.75 spec(b 3 )=3/3= 1.00 Thus, the specificity of b 3 (1.00) is more than specificity of a 3 (0.75), even though their depths are equal.…”

“…The natural language processing (NLP) literature includes a number of approaches to measure the semantic similarity, or more generally, relatedness between two terms or concepts [1][2][3][4][7][8][9][11][12][13][14][15]17]. Measures of semantic similarity and relatedness can be used in applications such as sense disambiguation, information extraction and retrieval, classification and ranking, detection of redundancy, and detection and correction of malapropisms.…”

Semantic Similarity Measure for Pairs of Short Biological Texts

“…Applications of this principle were implemented by Rada et al [3] in the MeSH ontology and by Caviades and Cimino [9] in the UMLS ontology. Variations of the path measure are those of Leacock and Chodorow [10] (lch), and that of Wu and Palmer [6] (wup).…”

A System for Ontology-Based Annotation of Biomedical Data

“…In the current prototype, concept recognition is done using a concept recognition tool developed by National Center for Integrative Biomedical Informatics (NCIBI) called mgrep. 10 We rely on this tool which reported a very high degree of accuracy (over 95%) in recognizing disease names [13]. The prototype design of the annotation level is such that we can plug-in other concept recognizers.…”