チームIrisapuのDRC2023プロジェクト説明
Team Irisapu Project Description for DRC2023
対話ロボットコンペティション2023の予選に向けて、GPT-3を活用し、シナリオを5つのサブシナリオに分割してプロンプトを作成することで柔軟な対話応答を生成するシステムを開発した。また、対話破綻に柔軟に対応する回復戦略を組み込んだ。
著者: Reon Ohashi, Shinjitsu Agatsuma, Kazuya Tsubokura, Yurie Iribe
分類: cs.RO, cs.AI
原文アブストラクト
This paper describes the dialog robot system designed by Team Irisapu for the preliminary round of the Dialogue Robot Competition 2023 (DRC2023). In order to generate dialogue responses flexibly while adhering to predetermined scenarios, we attempted to generate dialogue response sentences using OpenAI's GPT-3. We aimed to create a system that can appropriately respond to users by dividing the dialogue scenario into five sub-scenarios, and creating prompts for each sub-scenario. Also, we incorporated a recovery strategy that can handle dialogue breakdowns flexibly. Our research group has been working on research related to dialogue breakdown detection, and we incorporated our findings to date in this competition. As a result of the preliminary round, a bug in our system affected the outcome and we were not able to achieve a satisfactory result. However, in the evaluation category of "reliability of provided information", we ranked third among all teams.
