Reading papers on fake news detection (2): semi supervised learning and graph neural networks for fake news detection

Paper title ：Semi-Supervised Learning and Graph Neural Networks for Fake News Detection
Source of the paper ：IEEE 2019

# Based on news text 、# Semi supervision 、# The graph structure

One 、 The motivation of the article

1、 Data labels are usually very vague and sparse , Therefore, semi supervised detection method is selected ;
2、 The graph model is expressive , Be able to capture the context dependency between articles , So as to alleviate the problem of insufficient labels .

Two 、 frame

1、 Use word embedding to obtain the vector representation of the article in low dimension ;
2、 Capture the context similarity between articles through graph based representation ;
3、 Using graph neural network to carry out classification task on limited labeled data .

3、 ... and 、 Model composition

1、 Vector representation of the article ： Based on pre training Glove Word embedding , Calculate the average vector of words in the article ;
2、 Construction of similarity graph between articles ： The article is the node of the graph , according to embedding space Euclidean distance in search k-nearest neighbours, Build edge ;
3、 classification ： Two graph neural network methods are used ,GCN and AGNN（ Attention map neural network ）.

Four 、 experimental result

Data sets share 150 Tagged articles ,75 really ,75 false .

This paper is 2019 year ,（ One ） The paper is 2021 year , But in fact, it seems so ,2021 The paper in did not have much innovation , It only makes use of WMD Algorithm , And experiments are carried out on several public data sets , Good results have been achieved ？？