Text Classification Of Human-Computer Interacting Words Based On Deep Learning

With the rapid advancement of computer technology and artificial intelligence technology,more and more iintelligent products have entered people's lives,it is likely that human beings have entered the era of human-computer interaction.How to get useful information from the text of human-computer interaction and let the machine respond correctly has become the focus of scholars and commercial companies.This paper mainly studies the techniques of text classification in human-computer interaction field in the TCL company smart TV scene,and proposes a classification decision system which breaks the limitations of the original classification model that uses a single model,and this paper also carefully studies the hierarchiacal classification modual in the classification decision system.The main work of this paper is:1.First of all,this paper introduced the basic concepts of text classification,text language model,neural network language model,Word2 vec which is a word vector training tool and classification models commonly used in deep learning;secondly,use the k-means algorithm to cluster the human-computer interaction texts,preliminary determine of text data category;then according to the scene of the smart TV and the characteristics of the data,the boundary of the text in human-computer interaction field is determined,and the data module is divided;finally,according to the characteristics of the data module,the decision classification system is proposed.The experimental results show the accuracy of the text classification of human-computer interaction has reached 99.99% in the TCL company smart TV scene,the accuracy of classification has been improved by 19.99%-27.33% compared to textCNN model and textRNN model.2.This paper designes the convolutional neural network model to achieve the first-level classification in the hierarchical module using different sizes of convolution kernels which overcomes the disadvantage of the traditional machine learning that neglects the semantic relationship between words and words and falls into the local optimal shortcoming easily;then this paperuse the SoftMax classification function,in this way,the feature is extracted and compressed by continuous convolution layer and pooling layer,so that the extracted feature is compressed in the low-dimensional feature map;finally,the model is evaluated by the cross entropy loss function that is inversely adjusted by the ADMD algorithm to find the optimal parameters.The experimental results show that the accuracy rate of theconvolutional neural network is 99.91%.3.This paper designes a hidden layer with a single neuron's residual fully connected neural network that can increase the depth of the model without degrading the network.Moreover,the structure of the hidden layer with a single neuron can effectively reduce the complexity of the network,and the training speed of the model can be greatly improved.Set the TCE loss function so that the classifier can still maintain the training characteristics of the TCE loss function under no-noise conditions in the presence of noise.The experimental results show that the accuracy of the fully connected neural network based on the hidden layer with a single neuron is 99.99%.