| 研究課題/領域番号 |
18K19841
|
|---|
| 研究機関 | 京都大学 |
|---|
研究代表者 |
Adam Jatowt 京都大学, 情報学研究科, 特定准教授 (00415861)
|
|---|
| 研究期間 (年度) |
2018-06-29 – 2021-03-31
|
| キーワード | question answering / document archives |
|---|
| 研究実績の概要 |
Question answering system for event-related questions on long-term news article archive has been developed. The system uses additional temporal reranking module for searching for relevant documents by using their publication dates and content dates. We have also created a special test set containing 200 temporal questions to test the system. Bothe the methodology as well as the test set have been described in the publication in ECIR2020 conference. Moreover, that paper was selected for an extension in Information Retrieval journal following very good comments from the reviewers.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
Due to covid-19 pandemics we could not complete the research. The test set needs to be extended to have more test questions needed for verification of our proposal. Additionally, the current methodology works only for questions that deal with only single events and that have implicit temporal attributes. These need to be improved to make a robust system Further dissemination of the research results is also needed.
|
| 今後の研究の推進方策 |
We will extend the testset by manually creating or selecting possible questions for the document dataset that we use. Our method will be also extended to cover cases of questions that contain explicit dates in their content. We will also approach the task of answering questions that deal with several events. Lastly, we will seek venues for further research result dissemination.
|
| 次年度使用額が生じた理由 |
We will use the remaining budget for paying for attending the conference to disseminate research results. Additionally, some funds will be spent on crowdsourcing job for generating more test questions. Finally, we will purchase additional hardware such as GPUs to be able to conduct research when using large scale data.
|
