Assessing a BERT-based model for analyzing subjectivity and classifying academic articles

Atif Mehmood, Farah Shahid, Rizwan Khan, Shahzad Ahmed, Mostafa Mahmoud Ibrahim, Zhonglong Zheng

Abstract: The metaverse concept extends beyond virtual worlds and can be applied to collaborative analysis environments. Data analysts worldwide may read academic article extracts in real-time in a shared digital workplace to mine and analyze data using the metaverse. Furthermore, a semantic metaverse in natural language processing might also involve creating a digital environment with linguistic and semantic connections. Many dedicated researchers navigating the complexities of natural language processing in the metaverse era have spent considerable time searching for relevant papers. However, online reviews and evaluations of articles are helpful for their assistance and may save the researcher time. In this work, human specialists manually produced a dataset from four conferences and evaluated subjectively using rule-based techniques. Subsequently, we aim to evaluate the effectiveness of pre-trained word embeddings and pre-trained BERT models seamlessly integrated with convolutional neural networks. This endeavor focuses on the subjective analysis and classification of contributions and previous work sentences extracted from academic literature. For comparison, various deep learning architectures were systematically employed, including long short-term memory-GloVe and bi-directional long short-term memory-GloVe, alongside classical machine learning methods. Our findings show that the proposed BERT model achieved state-of-the-art performance in classification and subjective analysis tasks with an accuracy of 91.50% and F1 score of 91.00%. Finally, we plan to utilize sentence similarity to identify contributions within abstracts, thus outlining potential avenues for future research.

External IDs:dblp:journals/mta/MehmoodSKAIZ24