A Model Driven Framework for Ambiguity Detection in SRS

Abstract

The ambiguity detection in natural language processing (NLP) is critical to enhance the performance of the requirement based applications. There is lexical ambiguity, which is the complexity arising from confusing part words, syntactic, which is confusion about the structure of a sentence, and referential that happens when one has a problem identifying to which term a given term of reference is referring. This paper proposes a sound approach towards identifying these three categories of ambiguities using basic text processing and machine learning algorithmic analysis. The preparation phase itself entails significant pre-processing of the text in which we convert all text to lowercase, tokenize the text, eliminate the stop words and lemmatize the text. Such steps help to pre-process the text data to increase the chances of feature extraction as can be seen in this paper. In feature extraction the relevance of words is checked using Term Frequency-Inverse Document Frequency (TF-IDF) and syntactic feature extraction is done by Part-of-Speech (POS) tagging. Further, the next step involves the feature classification into ambiguous and non-ambiguous using various machine learning algorithms such as Logistic Regression, Random Forest and a Support Vector Machine (SVM). Indicator of a model’s accuracy include accuracy, precision, recall, F1 score and receiver operating characteristic-area under the curve (ROC-AUC). The findings are indicative of the value of combining more conventional methods of NLP with machine learning for improving the process of identifying levels of ambiguity. It is not only helpful in enhancing the explainability of automated systems but also, to a great extent, contributes to make NL Processing rather reliable and accurate in terms of the areas including but not limited to sentiment analysis, neural machine translation, and information retrieval.