Japan Journal of Medical Informatics
Online ISSN : 2188-8469
Print ISSN : 0289-8055
ISSN-L : 0289-8055
Original Article-Notes
Automatic Elusive Term Recognition in Functionality and Safety Information Targeted at General Consumers Using Several Glossaries
N TanemuraS KimuraY MachiiY ChoM KakizakiR OnoderaH Urushihara
Author information
JOURNAL FREE ACCESS

2019 Volume 39 Issue 2 Pages 45-60

Details
Abstract

 The subtle differences in the common ground of shared assumptions between consumers and professionals may cause miscommunication. To avoid it, it is crucial to identify technical words and make them consumer-friendly. However, there is no dictionary custom-made for identifying words commonly found in functionality and safety information on foods with health claims for general consumers. This study aims at creating a new dictionary for morphological analysis and evaluating its accuracy. We created 203,095 dictionaries initially, aggregating from ten glossaries consisting of medical or clinical trial terminology (total number of extracted terms: 973,895). We then randomly extracted 66 from 1,310 pieces for general consumers (4.65%) to evaluate the accuracy of the dictionary (data were collected in July 2018). Therefore, it was found that one of dictionaries, which had the highest recall, showed that the F value reached a maximum of 0.961 (number of detected cases were 529 from manual and 1,725 from the MeCab; the accuracy reflected a precision of 0.283; recall was 0.924; and F-value was 0.434). Additional empirical studies are needed to further validate the dictionary created in this study.

Content from these authors
© 2019 Japan Association for Medical Informatics
Next article
feedback
Top