| Text classification refers to associating a segment of text with one or more classification labels based on the content features of the text under fixed classification labels.For example,in the field of news,using text classification technology to process massive news texts can effectively organize and organize news texts,and to some extent,improve the phenomenon of information clutter in news texts.However,existing text classification methods generally face issues such as the inability to accurately define the semantic relationships between texts during text representation and the low efficiency of feature extraction.This thesis conducts research,analysis,system design,and implementation on existing text classification algorithms,and constructs a text classification system with high efficiency and accuracy.The main tasks are as follows:(1)A semantic fusion edge weight calculation method has been proposed.This thesis proposes an edge weight calculation method that integrates semantics to address the issue of neglecting the similarity and synonymy between words in existing methods for text vector representation.When calculating edge weights,the similarity between two nodes is first analyzed to determine whether they are similar.Then,the similarity factor of the nodes is calculated,and finally,the similarity factor is introduced into the edge weight calculation.And multi-dimensional experiments were conducted on the20 NG dataset to validate the algorithm.The results showed that compared to existing methods for calculating edge weights between nodes,this method has better performance in edge weight calculation and text feature extraction,effectively improving classification efficiency.(2)A text classification model that integrates attention mechanism into graph convolutional networks is proposed.In response to the problem of poor global feature extraction in existing text classification models,this thesis combines the graph convolutional network model with the multi head self-attention mechanism,and uses the edge weight calculation method proposed in this thesis to fuse semantics for text representation.The Attention Text Graph Convolutional Network(ATGCN)with attention mechanism is proposed for text classification tasks.To verify the effectiveness of this model,experiments were conducted on five representative text datasets: 20 NG,R8,R52,Ohsumed,and MR.The results showed that this model outperformed existing text classification models in terms of accuracy and efficiency.(3)We have designed and implemented a text classification system based on graph convolutional network fusion attention mechanism.In order to provide users with a good text classification environment,this thesis designs and implements a text classification system based on the proposed text classification model ATGCN.The system provides functions such as text classification,user information management,and result display.The system adopts the Spring boot architecture and is developed using Python language.VUE implements the front-end interface functions,and the database uses My SQL.The effectiveness and usability of the system have been proven through relevant testing. |
