Drug safety is an important science to detect threats related to medication consumption. Drug monitoring is often referred to as drug defence. For example, medicine's side effects can be caused by interactions, excessive doses and violence. In addition, information must be collected and mapped to predefined terms to find models in which the unintended effects are induced. This mapping is today manually conducted by experts, which can be very time consuming and challenging. In this paper the aim is to automate the mapping phase of side effects using techniques for machine learning.The model was created using data of pre-existing mappings of literal side effect expressions. The last design used the pre-trained BERT language model and the latest findings were obtained inside the NLP. The final model was correct at 80.21 percent in the evaluation of the test range. Some wordings were found to be very difficult to define for our model, mainly because of uncertainty or lack of literal knowledge. Since a threshold was introduced that left the most difficult to identify wordings for manual mapping, which is very important to make mapping correctly. However, this method could still be improved because suggested terms were created from the model, which could be used as support for the manual mapping specialist.

No abstract