Browsing by Author "1911110042"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Open Access
    Adaptative NLP for Low Resource Environment and SAFE: Semantic-Syntactic Based Text Augmentation Technique
    (North South University, 2023) Sunjare Zulfiker; S.M. Ashraful Hasan; Fahid Shadman Karim; Shahriar Karim; 1912050042; 1911570042; 1911110042
    Natural language processing (NLP) is a popular method for extracting context and syntax from text data for use in automation. NLP research focuses on dominant languages and advanced computing, neglecting low-resourced languages with limited computational capabilities. In this research, We propose the use of data augmentation (SAFE) approaches, suggest methods including Fusion architecture models (BiLSTM+CNN+BiLSTM) with optimized cyclic learning rate scheduler, layer-wise optimization, freezing early layers, and employ the multichannel approach to address the issue of low resource text data. In the domain of NLP where vast amounts of labeled data are scarce or expensive to collect, we offer our proposed method of SAFE augmentation technique for sentiment classification. This augmentation technique identifies the Subject, Verb, and Object in a sentence and alternates their position in six different permutations. The Object is also broken down into Subject, Verb, and Object again and is further rearranged. This aims to preserve the semantic meaning of the sentence while allowing some syntactic variation. We have also generated and calculated semantic and syntactic scores of the augmented sentences with respect to the original sentences.