2018 Fiscal Year Annual Research Report
Novel Technologies for Improving Comprehension and Utilization of Historical Knowledge
| Project/Area Number |
17H01828
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
Adam Jatowt 京都大学, 情報学研究科, 特定准教授 (00415861)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
澄川 靖信 首都大学東京, 大学教育センター, 助教 (70756303)
|
|---|
| Project Period (FY) |
2017-04-01 – 2021-03-31
|
| Keywords | news archives / digital history / computational history / summarization / entity similarity |
|---|
| Outline of Annual Research Achievements |
We have designed, implemented and tested a novel system for a challenging task of comparative summarization in news archives. Specifically, the proposed method applies Integer Linear Programming approach to connect past and present event-describing sentences in order to form a concise summary related to user query. This work appeared as full paper at WSDM2019 conference.
The next research was about finding similar terms from news archives to the input query from the present. We have applied neural network for this problem and developed demo system which was described as demo paper in WSDM2019 conference.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The current progress of the project is going well. There are no delays or any problems related to its execution.
We are working now on improving the summarization system as well as on work for estimating collective memories and historical references in Twitter. We are also proposing typicality based generation of pairs of similar objects: one object from the past and one from present. This approach allows to map a set of terms (e.g., celebrities) from the past with ones from the present.
|
| Strategy for Future Research Activity |
In future work we will provide novel techniques for suggesting, reformulating and extending user queries for historical news archives. First, we will provide taxonomy of search intents for document archives and then we will design effective methods towards supporting user search.
The other long-term research goal is to provide effective methods for extracting numerical information from the past by processing raw news data, such as changes in the value of some variables extracted from news archives over time.
|
