Quantum-inspired Model Based On Deep Neural Network For Sentiment Analysis

The era keeps advancing,and the technology also keeps advancing.Nowadays,mobile Internet has become more and more popular,and modern life is gradually entering the era of artificial intelligence.The Internet has become an important platform for the masses to express their opinions and suggestions in daily leisure or work and study.Therefore,a large amount of textual information is generated on each platform and in all walks of life,and the public sentiment lies in these opinions.At the same time,these textual emotions may have a certain degree of influence on the future development trend of the event.Sentiment analysis aims to judge the sentiment polarity of various types of texts at the document and sentence level.Both in theory and practice,the meaning is very important,and it is a hot topic in natural language processing.At present,the basic sentiment analysis methods widely used by the public,one is based on sentiment lexicon,and the other is based on machine learning.From some perspectives,these methods take into account some semantic information,but there are still no effective coding of semantic subspace mixing.Therefore,the feature interaction between emotional words is ignored.In addition,traditional machine learning methods need to rely on the effect of tagging artificial emotional words,and their scalability is poor.Deep learning methods largely ignore the capture of emotional information,but emotional information is very important in this task.In this article,we combine quantum mechanics,deep learning and natural language processing technologies,and propose a deep neural network sentiment analysis model based on quantum inspiration.They are the CNN-QI model and the CNN-BiLSTM-based on it.QI model,the model proposed in this paper combines the concept of density matrix in quantum mechanics into the neural network language model,and then applies it to sentiment analysis tasks.By treating each embedding vector as the observation state of each word,a sentence corresponds to a mixed state represented by a density matrix.The density matrix can encode more semantic dependencies,integrate it into the neural network,and pass the reverse The propagation algorithm automatically updates the model,which can effectively encode the mixture of semantic subspaces and reflect the degree of dispersion of sentence words in the embedded space,encode semantic dependence more effectively,and better capture the interaction of emotional characteristics between words Information and long-term dependence.Finally,experiments were carried out on the IMDB English data set and Weibo Chinese data set,which objectively proved the feasibility and effectiveness of our proposed model in sentiment analysis tasks.