Automated Plagiarism Detection Model Based On Deep Siamese Network

Abstract

This paper presents a novel deep Siamese network for automatic plagiarism detection. Our model utilizes a large-scale pre-trained model BERT (bidirectional encoder representations from transformers) to represent the text as word vector, and uses Bi-LSTM (bidirectional long short-term memory) net works to obtain the contextual semantic features of the text, and designs a text semantic interaction me chanism to obtain the interactive semantic features. Our model uses Siamese network to uniformly map matched text pairs into the same parameter matrix s pace. Meanwhile, our model uses multi-head self-attention to fuse text pair vectors for accurate semantic alignment and similarity measures. The experiment al results show that the effect of this model can identify and detect plagiarized text.