Topic model based recommendation system to identify operations that are missing in the treatment

In the medical field, although it is extremely important and a legal obligation to record the pro-cedures applied to patients by the health personnel, generally the operation lists are incomplete. Omissions in the operation lists can cause unexpected results for patients. In addition, inadmis-sible operation lists on billing operations applied to patients cause financial problems for both health institutions and patients because operation lists are used for invoicing process. There-fore, the main objective of this study is to develop an expert recommender system which can predict the omissions in the operation lists with a high success rate, which both threaten human health and cause economic problems for patients and medical centers. In this thesis study, we propose a new model different from the previous attempted solutions which tried to predict omissions in the operation lists using the Latent Dirichlet Allocation method, the proposed method uses the ICD-10 code as a new observed variable. The first experiments are carried out with Logistic Regression and Latent Dirichlet Allocation methods which had previously achieved success in this field. Precision, recall, F1 measure and MRR values are used as evalu-ation metrics, and the results of the proposed model with the Logistic regression method and the classical Latent Dirichlet Allocation method are compared based on the evaluation metrics. According to the results of the experiments conducted on three different datasets, it is ob-served that the Proposed Method is 5% more successful than the LDA method and 13% more successful than the Logistic Regression method.