Scientific Article Classification: Harnessing Hybrid Deep Learning Models for Knowledge Discovery

In an era defined by the explosive growth of the scientific literature, the imperative for effective methods of organizing, categorizing, and accessing scholarly articles has become paramount. This study addresses this crucial need by delving into the realm of scientific article classification, aiming to enhance accuracy through the innovative integration of hybrid deep neural networks. Focusing on domains including Computer Science, Mathematics, Physics, and Statistics, the research sought to improve categorization using advanced techniques. A comprehensive dataset of 20,006 abstracts was curated through rigorous data collection and preprocessing. Experimental models, spanning various RNN architectures and CNN-RNN based hybrids, were employed to assess the efficacy of the approach. The innovative integration of CNNs and RNNs pioneered new horizons in feature extraction. Key findings reveal the proficiency of hybrid models in capturing both local nuances and sequential dependencies within abstracts. Notably, the CNN-BiGRU model trained on Word2Vec embeddings exhibited highest F1 score of 91.32\%. Furthermore, comparative analysis between GloVe and Word2Vec embeddings underscored the pivotal role of embeddings in extracting semantic information for accurate classification.