A cross-domain recommender system through information transfer for medical diagnosis

The electronic diagnostic records of patients, primarily collected by hospitals, comprise valuable data for the development recommender systems to support physicians in predicting risks associated with various diseases. For some diseases, record are not sufficient train a prediction model generate recommendations; this is referred as sparsity problem. Cross-domain offer solution problem transferring knowledge from similar domain (source domain) modeling facilitate current (target domain). However, building cross-domain system medical diagnosis presents two challenges: (1) uncertain representations, such symptoms characterized interval numbers, often used records, and (2) given different feature spaces domains disparate because diseases only likely share few symptoms. This study addresses these challenges proposing system, named information transfer (ITMD), provide personalized recommendations disease risks. In ITMD, novel dissimilarity measurement was performed diagnosis, represented numbers. space alignment technique eliminated divergence caused between collective matrix factorization enabled source target domains. Experiments case using real-world demonstrated that ITMD outperforms four baselines improves accuracy patients determining final diagnosis.