A Fine-grained Multi-label Privacy Detection Model for Unstructured Data Based on BERT Pre-training
Yuchuan Hu, Bin Guo, Cheng Dai, Qin Zheng, Fangpeng Weng, Zhongwei Li · 2022
At present, a major bottleneck in the development of artificial intelligence is the difficulty of training data collection. Data holders are reluctant to share data because of concerns that sharing data may reveal private information. In order to solve this problem, this paper proposes a fine-grained multi-tag privacy detection model for unstructured data based on BERT pre-training. The model extracts the feature representation at the sentence level through BERT, and then allocates the weight to each sentence through the attention mechanism at the document level to obtain the feature representation of the document. Finally, combined with the calculation results of tag relevance based on GCN network, the result of multi-tag classification of unstructured text is obtained. This model achieves good results in the data set used in this paper.