Abstract

In this work, the authors identified opportunities for improvements in conversational systems. In order to solve the conversational systems learning problems, this investigation proposes a new architectural model for the conversational system “BRasa,” consisting of two subsystems. The first, “BRasa_Assistant,” is oriented to direct communication with users, and the second “BRasa_LDS” is oriented to conversational system learning inspired by Linguistic Data Summarization techniques. BRasa_LDS generates summaries in natural language, which incorporate new knowledge into the conversational system database. In addition, is proposed a system of indicators for the self-assessment of the human–computer interaction of the conversational system. In the analysis results section, three sets of tests were designed to measure the quality of conversational system responses. The proposal is validated based on the criteria applicability and adequacy of the conversational system responses. It is shown that the application of linguistic data summarization techniques for learning conversational systems improves the behavior of these systems significantly.