Text hashing by semantic information based on BERT model

Text hashing transforms a text into a binary hash code, and similar texts correspond to similar hash codes. By using text hashing, we can accelerate the similarity retrieval of text in big data environment and reduce the storage costs. In view of most existing text hashing methods having not integrated contextual semantic information into hash codes, a text hashing method based on BERT (Bidirectional Encoder Representation from Transformers) model is proposed in this paper. Firstly, feature sentences are selected from the text, then the BERT pre-training language model is used to get the feature vector of each feature sentence, next, the weighted feature vectors are summed and binarized to obtain the text hash code. Experimental results show that both the precision and the recall rate of the proposed method outperform other text hashing methods