Pattern Construction for Extracting Domain Terminology

Proceedings of Recent Advances in Natural Language Processing, pages 420–426, Hissar, Bulgaria, Sep 7–9 2015. Pattern Construction for Extracting Domain Terminology Yusney Marrero García Agrarian University of Havana, Cuba yusneym@unah.edu.cu Paloma Moreda Pozo University of Alicante, Spain moreda@dlsi.ua.es Rafael Muñoz Guillena University of Alicante, Spain rafael@dlsi.ua.es Abstract The extraction of domain terminology is a task that is increasingly used for different ap- plication processes of natural language such as the information recovery, the creation of spe- cialized corpus, question-answering systems, the creation of ontologies and the automatic classification of documents. This task of the extraction of domain terminology is generally performed by generating patterns. In literature we could find that the patterns which are used to extract such terminology often change from one domain to another, it means the interven- tion of human experts to the generation and validation of these patterns. This article deals with a methodology for automatic obtaining patterns (Basic Patterns and Definitory Verbal Patterns) for extracting domain terminology and minimizing the manual work of the ex- perts. The obtained methodology was eva- luated in the computer science domain obtain- ing a 97 percent in the case of the values of the basic patterns and a 98 percent of the definito- ry verbal patterns. Then the methodology was tested in three other domains with similar re- sults, Agricultural Engineering (a 96 percent of the basic patterns and a 97 percent of the definitory verbal patterns), Veterinary Medi- cine (98% of the basic pattern and the definito- ry verbal patterns) and Agronomy (96% of the basic pattern and the definitory verbal pat- terns), showing that methodology can be ap- plied in any specialty curriculum documents. 1 Introduction The extraction of terms that characterizes a document is a task of vital importance in the development of recovery systems and information extraction. It is very important to get the patterns that characterize these terms for the proper functioning of such systems. In the present systems of natural language processes, there is a tendency which minimizes the human labor, leaving the processing of the whole information to the system but the final validation made by experts remains irreplaceable in many cases. Sometimes the obtained patterns change from one domain to another, so there are some methods to minimize the human intervention. It would be a great step forward for the work of such systems. The research paper is organized as follows: after presenting the state of the art (Section 2), we present in Section 3 Pattern Generation Process, Selection of the corpus (Section 3.1), Definitory context (Section 3.2), Definitory verbal patterns (Section 3.3) and Our proposal (Section 3.4). Then, the processes of Evaluation – Analysis for the Computer Science domain (Section 4) and then the evaluation of the obtained methodology in other domains will be presented (Section 4.1) and the ending with conclusions and future work (section 5). 2 State of the Art Obtaining patterns for the extraction systems are a task whose success will depend on the correct operation of …

Content truncated to 500 words