Leveraging BERT-GRU Model for Drugs Interaction Relationship Extraction
Hang Shen, Xun Zhu, Hongtao Deng · 2023
In the medical field, drugs are often used in combination to improve their efficacy, but it is also common for drug interactions to produce adverse reactions that can put a patient’s health at risk and even cause serious side effects or toxic reactions. In order to minimise the risk of adverse reactions from drug interactions, doctors need to understand the properties of each drug and provide appropriate dosing instructions based on the patient’s individual circumstances. In order to allow physicians to quickly extract drug-drug relationships, drug-drug interaction relationship extraction techniques need to be studied and analysed to establish a medical drug information base to assist physicians in accessing important information about drug-drug relationships and avoiding adverse drug interactions. This paper proposes a new algorithm for drug interaction relationship extraction, with two main studies covering the following aspects: (1) To address the problems of multiple meanings of words in relationship extraction, high computational complexity, low training efficiency and low generalization ability, this paper proposes a new model BERT-GRU based on recurrent neural network and BERT. in order to eliminate the influence of Chinese word separation ambiguity on entity relationship classification, the BERT is introduced as an embedding layer to better obtain the contextual information of Chinese characters; then The long-distance dependencies of entities in the sentences are captured by gated loop units and output to the Sigmoid layer for classification, so as to obtain better entity relationship classification results. (2) By comparing and analysing the experimental results of the BERT-GRU model with other models, the BERT-GRU model showed better performance on the drug-interaction relationship extraction task, especially achieving higher F1 values for relationships of advice, mechanism and int types. The BERT-GRU model performed even better compared to several other common models. This suggests that the combination of BERT and GRU is very effective in the drug interaction relationship extraction task. In summary, through a series of experimental comparisons, the BERT-GRU model proposes in this paper can effectively solve the problems of inability to distinguish between polysemous words and low training efficiency, and provides a more effective method for drug interaction relationship extraction.