Research On Truth Discovery Algorithm Based On Open Source Information

In the era of Big Data,there are huge scale of heterogeneous data and true information mixed with false information on the Internet.People can get the information they want from various sources,but the open source data is distributed in long tails and the relationship between sources is implicit and complex.In the face of these problems,thesis focuses on the truth discovery of the internet open source data.Based on deep learning methods,unsupervised multi-source truth discovery model based on GCN and single-source truth discovery model based on external information fusion are proposed for unlabeled multi-source text data and single-source text data,respectively.And an internet information detection and discrimination system based on truth discovery is implemented.Thesis includes the following details.1.Unsupervised Multi-Source Truth Discovery Algorithm Based on Graph Convolutional Network.Considering the problem of ineffectively combining text semantics and the relationship between data sources for truth discovery,thesis proposes an unsupervised multi-source truth discovery model based on graph neural networks.Based on the text characterization method of smoothing inverse frequency,this model uses graph convolutional neural network for inter-source relationship mining and whole graph embedding method for truth estimation.Based on the comparison experiments of existing methods,it is shown that the whole graph classification method can effectively improve the accuracy of truth discovery.2.Single-Source Truth Discovery Algorithm Based on External Information Fusion.Considering the problem that the existing single-source text information truth discovery methods incorporate few external knowledge dimensions,thesis proposes a single-source truth discovery model based on external information fusion.The model incorporates two dimensions of external knowledge: news dissemination information and user relationship.The BERT model is used for text representation learning.The graph convolutional neural network and node embedding methods are used for analyzing the two types of external knowledge.Finally a multilayer perceptron is used for truth discrimination based on the learned results of the above analysis.The comparison experiments on the user preference perception model demonstrate the improvement of the truth discrimination results by adding external knowledge of user relationships.3.Truth Discovery Based Internrt Detection and Discrimination System.Based on the proposed truth discovery models,thesis implements an internet information detection and discrimination system using Python Flask framework,which provides effective support for other applications such as public opinion detection and false information discrimination.