Keyphrase extraction by the use of glove and ResNeXt optimized by enhanced human evolutionary optimization (EHEO) algorithm

Keyphrase extraction (KPE) is an essential process in natural language processing, facilitating the document content summarization for diverse uses like search engine optimization and information retrieval. Nevertheless, manual extraction can be labor-intensive, and automated techniques often face challenges in understanding contextual relationships within the text. This research introduces an innovative method that employs the ResNeXt neural network architecture, optimized by an enhanced human evolutionary optimization algorithm, and integrated with GloVe-100 word embeddings. The model was evaluated utilizing the KP20k dataset, a commonly utilized resource that includes approximately 500,000 scientific papers labeled with keyphrases, the Inspec database that comprises 2,000 English abstracts, and the SemEval-2010 benchmark dataset that comprises 244 research papers. The proposed model was evaluated against other advanced approaches, such as Convolutional Neural Network, k-Nearest Neighbors, Support Vector Machine, BERT, Convolutional Neural Network-BERT, and Gated Recurrent Unit, and outperformed them with recall, precision, and F1-score values of 98.81%, 98.67%, and 98.74%, respectively. It could achieve 96.54%, 96.32%, and 96.43% in precision, recall, and F1-score on the Inspect dataset, respectively. Moreover, the proposed model indicated remarkable performance over the SemEval-2010 dataset by achieving 97.32% precision, 97.81% recall, and an F1-score of 97.56%. These findings highlighted the efficacy of integrating high-quality embeddings and optimization strategies with advanced neural architectures. The suggested model provided a strong and effective solution for automatic extraction, which can be potentially used in various fields.