Medicine Recommendation for Allergic Rhinitis Based on Canonically Correlated Autoencoder

Abstract

In the electronic health records of allergic rhinitis patients, there are a large number of text-type chief complaint information, which contains key information of doctor making diagnosis and prescribing medication for patients. However, most of the existing medicine recommendation algorithms only focus on the use of numerical and structured data of patients. To solve this problem, this paper proposes a medicine recommendation algorithm for allergic rhinitis based on deep canonically correlated autoencoder. Firstly, this paper extracts symptom standard information from the chief complaint through a structured representation method based on search engine. After that, considering the strong correlation between patients' symptoms and medication, a deep canonically correlated autoencoder model is built to extract the features of the data and establish the correlation between chief complaint symptoms and medication. At last, Top-N recommendation for allergic rhinitis is made according to the symptom represen-tation and medication representation of patients. Experiments on a real electronic medical record dataset from the otolaryngology department of a first-class hospital demonstrate the accuracy and effectiveness of the algorithm.

Keywords

• medicine recommendation; deep canonically correlated autoencoder; allergic rhinitis; information extrac-tion; chief complaint