Страница публикации

Transformation of Decision Tables for the Creation of a Knowledge Base for Interpreting Facial Signs of Emotions

Авторы: Yurin A.Y., Dorodnykh N.O., Nikolaychuk O.A., Stolbov A.B.

Журнал: Proceedings 2022 Ural Symposium on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT 2022, Yekaterinburg, 19-21 September 2022)

Том:

Номер:

Год: 2022

Отчётный год: 2022

Издательство:

Местоположение издательства:

URL:

Проекты:

DOI: 10.1109/USBEREIT56278.2022.9923346

Аннотация: The paper describes the process of creating a knowledge base for the emotion signs interpretation module ("EmSi-Interpreter"), which is part of the software for the HR (human resources) manager support, namely "HR-Robot". The technology of prototyping of production expert systems based on transformations (PESoT) is used as a methodological basis. This technology has been expanded in the context of the current task of using a special type of decision table as a new source of information and defines the following steps for knowledge base engineering: analysis of domain and identification of the main entities and relationships; building decision tables; creating facts and rules; creating source code; testing and integration. Information from expert psychologists and elements of P. Ekman's theory is used to complete the decision tables. The evaluation of emotion recognition using the obtained knowledge bases was carried out on two data sets: a test video (studio recordings with actors clearly expressing emotions) and interviews with real respondents (shooting real persons using a laptop or a smartphone camera). In the first case, the score was 65%, in the second - 20%. The causes of the rather low scores are the following: the need to calibrate the system, i.e. adjust to a certain respondent in terms of the individual intensity of the manifestation of emotions; the need to improve video quality by minimizing shadows, glare, camera shake when shooting.

Индексируется WOS: Нет

Индексируется Scopus: Нет

Индексируется УБС: Нет

Индексируется РИНЦ: Нет

Индексируется ВАК: Нет

Индексируется CORE: Нет

Публикация в печати: 0