A Deep Learning Framework for Lexical Role Annotation in Persian Sentences Using CNN-BiLSTM and Evolutionary Algorithms

Natural Language Processing (NLP) is one of the fundamental branches of data science and artificial intelligence, aiming at the automatic analysis, understanding, and generation of human language. Recent advances in this field have played a significant role in improving text analysis systems, information retrieval, machine translation, speech recognition, and intelligent interactive systems. With the rapid growth of textual data and the increasing complexity of linguistic structures, traditional rule-based and simple statistical methods no longer provide sufficient capability for modeling complex and long-range linguistic dependencies. In recent years, deep neural networks, particularly hybrid architectures, have attracted considerable research attention. The combination of Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks enables the simultaneous extraction of local features and long-term temporal dependencies. However, the performance of such models is highly dependent on the proper selection of hyperparameters, and manual tuning is often time-consuming and suboptimal. Therefore, the use of metaheuristic algorithms for automatic hyperparameter optimization has emerged as an effective solution. In this study, a deep learning framework based on a CNN–BiLSTM architecture is presented, in which hyperparameter optimization is performed using the Cuckoo Search algorithm. Experimental results demonstrate that the proposed method achieves an accuracy of 93%, outperforming existing approaches and confirming its effectiveness in analyzing complex textual data while significantly reducing the error rate.