On Detecting Keywords for Concept Mapping in Plain Text

Abstract

The key terminology is very important for scientific works, especially for Natural Language Processing field. However, there is no optimal way to extract all the key terminology in a reliable manner. There by it is important to develop automatic methods for extracting key terms. This document presents a way to obtain the key terminology based on labels that were manually obtained by an expert in the area. Subsequently, we got POS (Part-of-the-speech) tags for each label, in which we obtained patterns from key terminology that were used as filters afterwards. Experiment 1 was tested using the labels obtained manually and the labels obtained by the proposed approach, with 60% of the corpus for training and 40% for tests. The patterns were evaluated with three different measures of evaluation such as precision, recall, and F-measure. Experiment 2 used three measures for ranking N-grams (sequence of terms), Point mutual information, Likelihood-ratio, and Chi-square. To obtain the best N-grams, we have implemented in experiment 3 intersections between the previous measures and filtering N-grams by POS patterns. Also, they were compared with the manually labelled set, evaluation measures were used to see its result, gave us a good recall moreover acceptable precision and F-measure. In experiment 4 POS patterns were tested in a much larger corpus of a different domain obtaining slightly higher results.