This study aimed to reach patients using different languages while providing an opportunity to enter symptoms in their everyday language text besides medical expressions of symptoms.

Named entity recognition (NER) techniques, based on natural language processing (NLP), were applied to develop a language independent predictive model. The research was based on extracting symptoms entered to the system by patient using NER method of NLP. In order to implement the system, python was used while pre-processing the data and string similarity function was used to estimate similarity with disease symptoms. Two sets were used for classification, one including only symptoms, and the other the matching diseases. Four thousand two hundred and eighty different symptoms were processed for the corresponding 880 diseases.

Each user symptom had a similarity score for each symptom in all diseases. Top N results with highest similarities were chosen from this list. The final N results are matched with diseases. According to these results, matched diseases were ordered in terms of the percentage of matched symptoms in the disease's symptoms. Extracted terms were implied as an input of the model and analysed for a matching diagnosis where an accuracy of 83 percent was accomplished when it is tested and compared using Mayo Clinic data for specific foreign languages other than English.

This language independent online diagnostic tool is a solution for both personal and clinical use and provides maintainable, updatable and more reliable diagnostics. This tool is particularly relevant today, with global mobility growing at a rate faster than the world`s population. We aim to upgrade the system by adding speech recognition and engaging it with the background (if available, electronic health records) of the patient.