An open-source MVP of the parser, which is already able to recognize the structure of molecules and determine what is a reagent and what is a product, was developed. At the stage of development, problems with recognition of some radicals and identification of images with poor resolution appeared. These problems will be solved by writing tests and improving the readout algorithms. In the future the possibility of implementing neural networks for recognition is being considered. Training of existing mechanisms on large and complex datasets is required in order to achieve maximum accuracy and high parser efficiency.
Лузанова А.М., Тонкий И.С., Головинский Р.П. (науч. рук. Скорб Е.В.) Parsing system for chemical reactions // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2023]. URL: https://kmu.itmo.ru/digests/article/10451