Multistream BertGCN for Sentiment Classification Based on Cross-Document Learning

Algorithm 1

The Algorithm of MS-BertGCN.

	Inactput: A dataset for sentiment classification.
	Output: Sentiment label.
(1)	Load the dataset;
(2)	Combine the documents in the training set based on within-class similarity;
(3)	Build heterogeneous graphs (including labeled data and unlabeled data and word nodes and document nodes) and initialize document nodes with BERT model;
(4)	Joint training of the BERT module and GCN module;
(5)	Use the trained BertGCN for inference;
(6)	Repeat 2–5 to fuse the predicted result.