An Entity Relationship Extraction Model Based on BERT-BLSTM-CRF Algorithm for Cosmetics Domain

In view of the lack of timeliness in the regulation of the cosmetics field, we establish a model for extracting the relationship between entities in the cosmetics field, assist the relevant departments in building an intelligent regulatory system, and rely on data to achieve scientific decision-making and effective regulation. In this paper, we use the BERT (Bidirectional Encoder Representations from Transformers) network model to train the word vectors, and combine the BLSTM (bidirectional long short-term memory) network with the fused positional attention mechanism to perform entity relationship extraction, and the extracted relationship features extracted by the BERT neural network are incorporated into the word dimension text vector, and then the entity pairs are extracted by the BLSTM with the fused positional attention mechanism, and finally the predicted labels are decoded by CRF (Conditional Random Field). The experimental results show that the BERT-BLSTM-CRF-based entity relationship extraction model in cosmetics domain constructed in this paper exhibits good feature extraction and classification performance.