| 研究課題/領域番号 |
19K12132
|
|---|
| 研究機関 | 大学共同利用機関法人情報・システム研究機構(機構本部施設等) |
|---|
研究代表者 |
金 進東 大学共同利用機関法人情報・システム研究機構(機構本部施設等), データサイエンス共同利用基盤施設, 特任准教授 (40536893)
|
|---|
| 研究期間 (年度) |
2019-04-01 – 2023-03-31
|
| キーワード | dialog / intelligent agent / natural language query / database search / task-oriented dialog / intent detection |
|---|
| 研究実績の概要 |
The research on search-oriented dialog systems for data science has been conducted so far for 3 years. While the total research period was originally planned to be 3 years, due to the pandemic situation, it was extended to 4 years. The research schedule was also changed. For example, the plan for organizing an annotation campaign for intent annotation was cancelled. Instead a pseudo annotation for intent detection was developed through a semi-automatic way. Following is overall achievements we have made so far:
(1) The user interface for developing dialog corpora was developed based on two open-source systems, PubAnnotation and TextAE; (2) The corpus of search-oriented dialogs was developed; (3) Annotation guidelines for intent annotation was developed; (4) Pseudo annotation for dialog intent detection was developed in a semi-automatic way; (5) intent detection system was developed by fine tuning the Transformer pre-trained language model; and (6) A prototype system of Search-Oriented dialog was developed.
In the last year of the project, (1) the pseudo annotation for dialog intent detection will get verified by human users, (2) A thorough evaluation on the search-oriented dialog system will be conducted, and (3) papers will be published based on the outcomes of the research.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Due to the pandemic situation the original schedule of the research has been changed substantially. However, thanks to the one-year extension which was generously granted by JSPS, the research is now in the stage of finalization.
|
| 今後の研究の推進方策 |
In the last year of the project, (1) the pseudo annotation for dialog intent detection will get verified by human users; (2) A thorough evaluation on the search-oriented dialog system will be conducted; and (3) papers will be published based on the outcomes of the research.
As the pandemic situation is still on-going, the user verification will be conducted using a crowd-sourcing platform. For paper publication, we are considering ACL and AAAI as prospective venues.
|
| 次年度使用額が生じた理由 |
Due to the pandemic situation, the two tasks, (1) the annotation of search intents to the dialog corpus, and (2) the user evaluation of the search-oriented dialog system, which were originally planned to be organized in the year R3, are delayed to happen in the year R4. Accordingly, we need funding for purchasing necessary devices, and for compensating the human labor for the annotation and the evaluation jobs. Also, need funding for preparing various hardware devices to be used for presentation, as well as for publication of the output of the research, e.g., conference registration and article processing fee.
|
