| In recent years,Internet technology has been presented to the public at a rapid development speed,and the amount of information it carries has soared year by year with the development of new technologies in big data and artificial intelligence.In the main dissemination medium of Internet information,short text messages generated by the explosive growth every day account for a large proportion.However,due to the sparse features and weak context of short texts,traditional text classification methods are becoming more and more difficult for short text information extraction.The deep learning theory has attracted the attention of scholars since it was proposed.The most representative one is the application of convolutional neural network models in text classification tasks.This model solves the problem that traditional text classification methods lack the ability to automatically extract text features,and shows good application value and development prospects.Therefore,this paper combines the convolutional neural network model to complete the following research work for short text classification:First of all,under the research background of short text classification methods at home and abroad,this paper deeply researches traditional machine learning text classification methods and the basic theories related to short text classification,and analyzes the current short text classification methods that can be used for reference and need to be improved.Secondly,in view of the traditional multi-size filter convolutional neural network in the text classification process,only simple word vector features can be obtained and the problem of important features of text context is ignored,this paper proposes a short text classification method based on n-gram and convolutional neural network.The short text classification method is based on a multi-size filter convolutional neural network,and uses the sliding window mechanism of the n-gram model to obtain the contextual relationship of these short texts.At the same time,in order to extract the key features of the short text,the attention mechanism is applied to the classification model,and the operation of combining two different pooling methods is used to make the classification result more accurate and reliable.The experimental results prove that the short text classification method based on N-Gram and convolutional neural network proposed in this paper has good classification effect and performance.Finally,the traditional short text classification method is facing the problem of poor classification performance when the short text data is sparse and the semantic features are insufficient.This paper proposes a short text classification method based on convolutional neural network and semantic expansion.First,the coverage of the pre-training word vector table is improved by finding possible spelling errors in the short text preprocessing process.Then,aiming at the problem of limited semantic information provided by short texts,the attention mechanism is used to find relevant words in short texts,and an external knowledge base is introduced to conceptualize short texts and related words,which expanded the semantics of short texts.Finally,the method combines the multi-size filter convolutional neural network model to extract short text features and complete the classification process.Experimental results show that the method is feasible in short text classification tasks,and the classification effect is significantly improved.There are 13 pictures,4 tables,and 56 references. |
