IADR Abstract Archives
Automatic Extraction Of Features From Unstructured Dentistry Referrals.
Objectives: To extract keywords from unstructured text of a database of referrals from the Clínica Odontológica UNAB (COUNAB), between 2017 and 2019.
Methods: COUNAB referrals were extracted from the U-Smile platform. Each referral contains information regarding administrative data, the specialty to be referred and the observation which is a narrative describing the reason for referral. The observation of each of the referrals was analyzed, which is in the form of unstructured free-text. The TF-IDF (Term Frequency - Inverse Document Frequency) method was used for the weighting of the importance of each word in the referral. This method returns a numerical value describing the level of relevance of each word to differentiate the referral from the rest, utilizing the frequency of the word in the referral and frequency of the word on the entire dataset to perform the relevance weighting.
Results: 924 referrals were obtained. The specialty with more referrals was Endodontics with 245 referrals. The keywords from each one of the referrals were obtained. The words with lower IDF values (providing less information in the referral) were patient, treatment and tooth, and those with higher IDF were nocturnal, neurological and complication.
Example of an automatic keyword (*) extraction using our proposed method: I am requesting evaluation and treatment possibility for partially edentulous* patient. History of hypertension* and diabetes*.
Conclusions: The important words retrieved by the method were the determinants to establish differences between referrals. With this method it is possible to explore unstructured data intuitively to extract important information in a more efficient way. This tool would help students and professionals in the selection of patients based on a given query, lowering the time consumed in searching patients. Using the TF-IDF method, the most important words were extracted within the observation attribute of each of the referrals.
Methods: COUNAB referrals were extracted from the U-Smile platform. Each referral contains information regarding administrative data, the specialty to be referred and the observation which is a narrative describing the reason for referral. The observation of each of the referrals was analyzed, which is in the form of unstructured free-text. The TF-IDF (Term Frequency - Inverse Document Frequency) method was used for the weighting of the importance of each word in the referral. This method returns a numerical value describing the level of relevance of each word to differentiate the referral from the rest, utilizing the frequency of the word in the referral and frequency of the word on the entire dataset to perform the relevance weighting.
Results: 924 referrals were obtained. The specialty with more referrals was Endodontics with 245 referrals. The keywords from each one of the referrals were obtained. The words with lower IDF values (providing less information in the referral) were patient, treatment and tooth, and those with higher IDF were nocturnal, neurological and complication.
Example of an automatic keyword (*) extraction using our proposed method: I am requesting evaluation and treatment possibility for partially edentulous* patient. History of hypertension* and diabetes*.
Conclusions: The important words retrieved by the method were the determinants to establish differences between referrals. With this method it is possible to explore unstructured data intuitively to extract important information in a more efficient way. This tool would help students and professionals in the selection of patients based on a given query, lowering the time consumed in searching patients. Using the TF-IDF method, the most important words were extracted within the observation attribute of each of the referrals.