2023 Fiscal Year Annual Research Report
Search-Oriented Dialog System for Data Science
| Project/Area Number |
19K12132
|
|---|
| Research Institution | 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) |
|---|
Principal Investigator |
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)
|
|---|
| Project Period (FY) |
2019-04-01 – 2024-03-31
|
| Keywords | Dialog Interface / Large Language Model / ChatGPT |
|---|
| Outline of Annual Research Achievements |
The project has been significantly impacted by two global events. Firstly, the Covid-19 pandemic restricted our ability to interact with research collaborators. Secondly, the advent of LLMs transformed the research landscape, necessitating adjustments to our approaches. In response, we decided to leverage the power of LLMs and focused on developing customization of LLMs.
The primary outputs of this research are the dialog interface implemented in LODQA (https://lodqa.org), as well as AnatomyMapper and BiomedicalContentExplorer, customized GPTs developed for searching biomedical databases (available from the GPT store). Additional outputs include a web-based user interface for analyzing and annotating dialogs, implemented as a feature of TextAE (https://textae.pubannotation.org).
Performance evaluation was successfully conducted in collaboration with participants of the 8th Biomedical Linked Annotation Hackathon (BLAH8). AnatomyMapper is currently being used by the developers of BodyParts3D, a database of anatomical 3D models.
Due to the significant shift in our research direction, we had to focus on development and performance evaluation until the end of the research period, which prevented us from completing the planned paper publications. However, one paper has been submitted to the journal Genomics & Informatics and is currently under review.
Overall, we have demonstrated that search-oriented dialog systems can be effectively developed through the deliberate customization of LLMs. We will continue publishing our results in journals and conference papers.
|
