A topic-aware classifier based on a hybrid quantum-classical model

Abstract In the era of Large Language Models, there is still potential for improvement in current Natural Processing (NLP) methods terms verifiability and consistency. NLP classical approaches are computationally expensive due to their high-power consumption, computing power, storage requirements. Another efficient approach categorical quantum mechanics, which combines grammatical structure individual word meaning deduce sentence meaning. As both theory natural language use vector space describe states more on hardware, QNLP models can achieve up quadratic speedup over direct calculation methods. recent years, significant progress utilizing features such as superposition entanglement represent linguistic hardware. Earlier research work has already demonstrated QNLP’s advantage speeding search, enhancing classification tasks’ accuracy providing an exponentially large state complex structures be efficiently embedded. this work, a model used determine if two sentences related same topic or not. By comparing our tensor network-based one, improved training by 45% validation 35%, respectively. The convergence also studied when varying: first, problem size, second, parametrized circuits model’s training, last, backend simulator noise model. experimental results show that strongly entangled ansatz designs result fastest convergence.