A Quantum-Probability-Inspired Complex-Valued Model for Multilingual Stance Detection

In this study, we propose a quantum-probability-inspired complex-valued model for multilingual stance detection. The model brings together ideas from granular computing and quantum theory to better capture semantic meaning across different languages. The proposed model combines contextual embeddings, graph convolutional networks, and a quantum-inspired feature interaction module (QFIM) to capture complex, high-order, and non-linear relationships in multilingual data. The QFIM models quantum amplitude-like interactions to represent overlapping semantic patterns in the latent space. This design helps the system better distinguish subtle differences in stance expressions. To strengthen the representation, a granulation mechanism based on contextual embedding similarity is employed to extract semantically coherent text granules. These granules expand the feature space and help align different elements of the stance structure more accurately. Experimental results on standard benchmark datasets show that the proposed model consistently achieves better performance than existing state-of-the-art methods.