Adverse Drug Reactions Recognition Based On Social Media

Along with the people living standard improvement,medical and health problems have been paid more and more attention by people,the safety of drugs is one of them.A number of adverse reactions caused by taking drugs occur frequently,even threaten life in serious conditions,so the adverse drug reactions have been widely concerned.With the popularization of the Internet,many websites concerned about medical and health have come into being.People can share and exchange medical experiences on these social media websites.User generated information contains a wealth of drug-related knowledge,therefore,social media has gradually become a quick and effective way to mine adverse drug reactions.Adverse drug reactions recognition is a special entity recognition task.Compared with traditional entities,adverse drug reactions contain both entities composed of continuous words(called “continuous entities”)and entities composed of discontinuous words or fragments(called “discontinuous entities”).For adverse drug reactions recognition task based on social media,on the one hand,this paper has been engaged in the research of the entity representation method which can represent both continuous and discontinuous adverse drug reactions.On the other hand,according to the characteristics of subject clarity about social media and strong field about adverse drug reactions,two methods based on conditional random field and deep learning are proposed to recognize adverse drug reactions.Finally,the effectiveness of the proposed method is verified by experiments.This paper adopts two kinds of methods can simultaneously represent continuous and discontinuous adverse drug reactions: BIOHD and Multi-label,which recognize adverse drug reactions based on social media under the framework of conditional random field and bidirectional LSTM-CRF model.To improve the performance of adverse drug reactions recognition,we integrate subject features and knowledge base features.The effectiveness of the proposed method is verified on an open corpus about adverse drug reactions based on social media.The experimental results show that Multi-label is superior to BIOHD on entity representation method about adverse drug reactions based on social media.The bidirectional LSTM-CRF deep learning method proposed in this paper is better than conditional random field method in adverse drug reactions recognition based on social media.At the same time,the fusion subject and knowledge base features on the basis of bidirectional LSTM-CRF deep learning method have significantly improved the performance of adverse drug reactions recognition.