Abstract
This article sketches the development of a method for mining concepts applied on medical corpora in Spanish. Such method is based in the approach formulated by Ananiadou and McNaught, who give a special relevance to the need to create and use natural language processing (NLP) tools, in order to extract information from large collections of documents, such as PubMed (www.ncbi.nlm.nih.gov/pubmed/). Thanks to this repository, projects such as the Corpus Genia (www.geniaproject.org), the MEDIE search engine (www.nactem.ac.uk/medie/), which considers syntactic criteria and semantics to extract medical concepts, or the Open Biological and Biomedical Ontology Project (http://obofoundry.org/), which focuses on the development of ontologies that provide an organized knowledge system in biomedicine. Particularly, this proposal focused in two objectives: (1) the extraction of specialized terms and (2) the identification of lexical-semantic relationships, in concrete hyponymy/hypernymy and meronymy.
TopIntroduction
In recent years, the automatic processing of biomedical information has been benefited from advances made by data and text mining. An example of this advance is the book edited by Ananiadou & McNaught (2006), who give a special relevance to create and use tools capable to extract information from large collections of documents, particularly PubMed (http://obofoundry.org/), focused on the development of ontologies that provide an organized knowledge system in biomedicine.
In line with these projects, it is exposed here a method for performing a concept mining on biomedical documents in Spanish. This model is based on the automatic extraction of definitional contexts (or DCs), according to the framework developed by Sierra et al. (2008), Sierra (2009), Aguilar & Acosta (2016), as well as Aguilar et al. (2016). Our model has been sketched having in mind the following objectives:
- •
The linguistic analysis of definitional contexts identified in biomedical texts in Spanish.
- •
The creation of a linguistic corpus in Spanish that is representative for the biomedical area.
- •
The use of stochastic methods in order to provide empirical evidence to validate in linguistic analysis previously performed.
These objectives allow to establish a set of concrete tasks that can be implemented as modules for a mining system, concretely:
- a)
Extraction of terminological information: this one focuses on establishing a chain of text processing, which considers: (i) the selection, tokenization and syntactic annotation of a Spanish corpus in medicine; (ii) the identification of uni/multiword terms using a hybrid method (Acosta, Aguilar & Infante, 2015).
- b)
Identification of lexical relations: taking advantage of the term extraction and linking them to their definitions, it is possible to implement an ontology in Spanish, considering the recognition of lexical-semantic relations, specifically hyponymy-hyperonymy and meronymy, based on the proposal of Arp, Smith & Spear (2015).
The organization of this article is as follows: section 1 shows a general background behind the notion of concept mining; the section 2 describes how is performed the identification of DCs, in order to detect and extract terms that function as hyponyms, hypernyms and meronyms. The section 3 exposes the results of these extractions, delineating a possible ontology that organizes such lexical-semantic information. To conclude, the section 4 offers a summary along with a brief discussion respect to future applications of this model of mining concepts.
Key Terms in this Chapter
Lexical Relationship: A conceptual link established between words grouped around a lexicon.
Text Mining: A process of deriving high-quality information from text. This kind of task involves information retrieval, lexical analysis to study word frequency distributions, pattern recognition, tagging/annotation, information extraction, data mining techniques including link and association analysis, visualization, and predictive analytics. The overarching goal is, essentially, to turn text into data for analysis, via application of natural language processing (NLP) and analytical methods.
Concept: A mental construction which contains information about a specific entity based on our knowledge and our experience.
Hyponymy-Hypernymy: A kind of hierarchical lexical relation where is established a conceptual subordination between the meaning of two or more words (e.g., a robin [hyponym] is a kind of bird [hypernym]).
Meronymy: A semantic relation specific to linguistics, distinct from the similar metonymy. A meronym denotes a constituent part of, or a member of something (e.g., wheels are part of an automobile ; a policeman is a part of a police department , etc.).
Ontology: A system of representation that shows the formal naming and definitions of the categories associated to entities. It describes the properties and relations between the concepts refer by such entities, situated in a specific or general knowledge domain.
Term: A word or compound words that designates a specific entity, event or relation into the framework of a scientific or technical knowledge domain.
Concept Mining: A kind of task derived from the knowledge discovery, focused to recognize patterns in large collections of unstructured data (that is, linguistic corpora), which contain relevant information associated to concepts. For solving such task, it is employed different methods based on NLP (e.g., tokenization, lemmatization, POS tagging, syntactic and semantic analysis, and others).
Definitional Context: A discursive fragment constituted by a term, a definition, and linguistic or metalinguistic connector as predicative phrases, typographical markers, or pragmatic patterns.