| The development of quantum theory has led to the research of constructing language model based on quantum probability.Some researchers have proposed quantum language model(QLM)and applied it to information retrieval tasks.The model uses the density matrix of quantum theory to model the term dependence.Recently,in order to expand the theory and application of quantum language model,the density matrix is embedded into the neural network model,which uses the neural network technology to train density matrix.However,these two quantum language models still have some limitations: first,the quantum language model can not construct the interaction of different semantics between multiple words in a sentence;second,the neural network in NNQLM is mainly for improving the training effect,and the relationship between the neural network and the quantum language model needs further theoretical study.To solve these two problems,in this paper,we propose a language model based on quantum many-body wave function(QMWF-LM).In this model,the sentence is represented by the high-order tensor.We think that higher-order tensors have stronger expression than density matrices and can construct different semantic interactions of multiple words in a sentence.In order to make QMWF-LM applicable to practical natural language processing tasks,we explored the connection between high-order tensor decomposition and convolutional neural network(CNN)and solved high-order tensors based on CNN model.QMWF-LM construct the relationship among the CNN,language model and quantum theory.In order to verify the validity of QMWF-LM model,we design an end-to-end algorithm and apply it to question and answer(QA)tasks.The results on three traditional QA datasets TREC,WIKI and Yahoo based on community show that QMWF-LM is significantly improved the previous quantum language model. |
