Hybrid Quantum–Classical Modeling for Persian Poetic Rhyme Detection

Detecting rhyme in Persian poetry presents a computational challenge due to the frequent divergence between its orthographic representation and phonological realization. While deep learning models can address this ambiguity, they typically require considerable computational resources and large parameter counts. To mitigate this issue, we propose CRNN-VQC, a hybrid architecture that integrates Convolutional-Recurrent Neural Networks (CNN–BiGRU) with Variational Quantum Circuits (VQCs). Instead of standard dense layers, the model embeds character-level features into an ۸-qubit parameterized quantum circuit via a multi-stage data re-uploading scheme, aiming to facilitate non-linear mapping within a high-dimensional quantum space. Evaluated on a curated corpus of ۲۰,۱۷۱ hemistich pairs, CRNN-VQC achieves a test accuracy of ۹۶.۹۵% and an F۱-score of ۹۶.۳۰%, statistically outperforming equivalent classical baselines. Importantly, the proposed model achieves competitive performance while utilizing approximately three orders of magnitude fewer parameters compared to large-scale pre-trained language models. Additionally, robustness tests under simulated quantum noise indicate the model’s stability, suggesting its potential viability for Noisy Intermediate-Scale Quantum (NISQ) devices. These findings suggest that hybrid quantum–classical frameworks may offer a highly parameter-efficient alternative for analyzing the phonology of morphologically rich languages.

کلیدواژه ها:

Quantum Machine Learning ، Hybrid neural networks ، Persian Poetry Analysis ، Rhyme Detection ، Variational Quantum Circuits