2020 Fiscal Year Final Research Report
Large-scale conversation corpus creation by automatic quality estimation
Project/Area Number |
18K11435
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 61030:Intelligent informatics-related
|
Research Institution | Osaka University |
Principal Investigator |
Arase Yuki 大阪大学, 情報科学研究科, 准教授 (00747165)
|
Project Period (FY) |
2018-04-01 – 2021-03-31
|
Keywords | 対話システム / 応答生成 / ニューラル対話モデル / 対話コーパス |
Outline of Final Research Achievements |
Neural conversation models based on deep neural networks have significantly improved fluency in response generation. However, these fluent responses are not necessarily satisfactory. Previous studies have revealed that automatically generated responses would break down a conversation between a user and system. Furthermore, even though these automatic responses are acceptable, they tend to be less attractive to users and may degrade user engagements. To address these problems, we developed novel neural conversation models that are sensitive to users’ utterances and generate meaningful responses. Further, we created a conversation corpus that opens up a new door to the conversational system’s research. It contributes to advancing natural language understanding technology to infer users’ hidden intents from their indirect utterances.
|
Free Research Field |
自然言語処理
|
Academic Significance and Societal Importance of the Research Achievements |
本研究で開発したユーザの発話内容に則しつつ情報量の多い応答を生成できるモデルは、これまで当該研究分野で広く認識されていた問題を解決するものであり、学術的だけでなく産業的貢献も大きい。また本研究で構築した対話コーパスは、これまで重要性を認知されながら手つかずであった、ユーザ発話に隠された言外の意図の推定を可能とするものであり、対話システム研究に新たな扉を開く、顕著な学術的貢献を持つ。
|