Comparative Analysis of Approaches to Building Medical Dialog Systems in Russian

Nowadays dialog systems have great promise in the field of medicine and healthcare. Only a few medicine dialog systems presented in the literature are accompanied by an experimental effectiveness evaluations. Moreover, they cover only English-speaken context. In this paper we set the task to conduct a comparative analysis of the effectiveness of Russian-language mixed-initiative medical dialog systems, depending on the chosen architecture and method of processing the users’ intentions. We have developed and compared three types of chat-bots; Frame-based, ML-based and Ontology-based. As the metrics used the accuracy of the intent recognition as the percentage of intents correctly defined by the system relative to the total number of processed utterances. The results show that the accuracy of the intent recognition in all three approaches is quite the same, so finally we propose the architecture of s combined dialog system which covers all the needs for Russian-language medical domain.