News text classification method based on graph network pooling
A text classification and pooling technology, applied in text database clustering/classification, unstructured text data retrieval, biological neural network models, etc., can solve the problem of not calculating soft edge weights, node features are too smooth, and information loss is much, etc. problem, to achieve the effect of accurate calculation
Active Publication Date: 2021-03-12
HANGZHOU DIANZI UNIV
View PDF6 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
However, it neither aggregates node information nor calculates soft edge weights, so it cannot effectively preserve node and edge information
ASAP has improved this, but it is easy to cause node features to be too smooth when aggregating node information, and more information is lost, so the news text classification effect is not good.
At present, there is no pooling method to avoid the problem of over-smoothing of node features while retaining node information and edge information in the graph
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0052] This embodiment provides a news text classification method based on graph network pooling, such as figure 1 shown, including steps:
[0053] S1. Combine structural information and feature information in the attention mechanism, and calculate the similarity score between nodes in the first-order neighborhood in the graph neural network, and obtain an attention mechanism with similar nodes;
[0054] S2. Use the sparse probability activation function sparsemax algorithm to sparse the obtained attention mechanism to obtain the clusters corresponding to the nodes;
[0055] S3. Use local aggregation convolution to calculate the score of each cluster, and judge the amount of information contained in the cluster by the score;
[0056] S4. Use topk to select the top scorer clusters, and reconnect the selected clusters to obtain the final pooled neural network.
[0057] This embodiment is specifically as follows: firstly, in the graph, nodes with high similarity are adaptivel...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a news text classification method based on graph network pooling, and the method comprises the steps: S1, combining structural information with feature information in an attention mechanism, calculating a similarity score between nodes in a first-order neighborhood in a graph neural network, and obtaining an attention mechanism with similarity nodes; S2, performing sparsification on the obtained attention mechanism by adopting a sparse probability activation function sparsemax algorithm to obtain a cluster corresponding to the node; S3, calculating the score of each cluster by adopting local aggregation convolution, and judging the amount of information contained in the clusters according to the scores; and S4, selecting a previous cluster with the highest score byadopting topk, and carrying out edge connection on the selected cluster again to obtain a final pooled neural network.
Description
technical field [0001] The invention relates to the technical field of news text classification, in particular to a news text classification method based on graph network pooling. Background technique [0002] With the rapid development of the era of big data, text data on the Internet has shown explosive growth, and it is of great significance to mine effective information from massive data. Since news texts have no fixed format, are diverse, and update quickly, traditional manual classification is inefficient and subjective. The graph neural network is introduced into the classification of news articles. The news text is regarded as a graph, and the nodes in the graph are composed of words. News text classification mainly focuses on the overall characteristics of the text, that is, the research object is the whole graph itself. Graph neural networks are generally composed of convolutional layers and pooling layers. The research on graph convolution is very rich, and its ...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More IPC IPC(8): G06F16/35G06F40/194G06N3/04G06K9/62
CPCG06F16/35G06F40/194G06N3/045G06F18/22
Inventor 朱小草郭春生陈华华应娜
Owner HANGZHOU DIANZI UNIV