| In linguistics,the uncertainty of context caused by the phe-nomenon of polysemy is widespread,which attracts much attention.A word containing multiple meanings could correspond to a single quantum particle which may exist in several possible states,and a sentence could be analogous to the quantum system where particles interfere with each other.Quantum-inspired complex word embedding based on Hilbert space plays an important role in natural language processing(NLP),which fully lever-ages the similarity between quantum states and word tokens.In addition,quantum-inspired neural networks based on complex word embedding in-cluding CE-Sup and CE-Mix have been applied to text classification tasks in natural language processing.However,the sentence-level density matrices used to represent text data in the above models are just a linear superposi-tion of the word-level density matrices,lacking the sequence information of text sentences.Therefore,motivated by quantum-inspired complex word embedding,we combine quantum theory with deep learning technology,inject quantum characteristic into neural network models,thereby improve the interpretability of neural networks and enhance the feature extraction ability of the network models by deep learning technology,implement the models under the open source deep learning platform and apply the models to Chinese and English text classification tasks in natural language pro-cessing.Based on the above,the main work of this paper is summarized as follows:Firstly,two quantum-inspired deep neural networks called ICWE-QNN and CICWE-QNN are proposed to complete the text binary classification tasks in natural language processing with the help of combining the the-ory of quantum mechanics with deep learning technology,basing on the similarity between the superposition of quantum states and the polysemy phenomenon of word tokens.ICW-QNN can avoid random combinations of words,while CICWE-QNN fully considers the textual features of the result matrix on the basis of ICWE-QNN.Secondly,the two quantum-inspired deep neural networks proposed in this paper,ICWE-QNN and CICWE-QNN,are implemented in the two open source deep learning platforms Py Torch and Paddle Paddle,and we compare the performance difference of the proposed models when the pro-posed models are running on the both platforms.Thirdly,experiments verifying ICWE-QNN and CICWE-QNN have been conducted on 3 open source Chinese and 5 benchmarking English datasets~2.The experimental results show that the both quantum-inspired deep neural networks in this paper have great performance on Chinese and English classification tasks.For instance,compared with classical mod-els Caption Rep BOW,Dict Rep BOW and Paragram-Phrase as well as pro-posed quantum-inspired neural networks including CE-Sup and CE-Mix,our model CICWE-QNN achieves the highest classification accuracy on four English datasets including SST,SUBJ,CR and MPQA,up to 85.0%,93.2%,83.3%and 87.2%respectively. |
