2018 Fiscal Year Annual Research Report
Framework for context-sensitive fact extraction over web data.
| Project/Area Number |
17K12786
|
|---|
| Research Institution | National Institute of Advanced Industrial Science and Technology |
|---|
Principal Investigator |
LEBLAY JULIEN 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究員 (70757377)
|
|---|
| Project Period (FY) |
2017-04-01 – 2019-03-31
|
| Keywords | knowledge graphs / context-dependency / machine learning |
|---|
| Outline of Annual Research Achievements |
During the first period of this grant, we had pursued and extended preliminary work on context-dependent reasoning. In particular, we developed a usable platform to demonstrate how the veracity of certain claims vary with respect to context given a set of, possible incomplete or uncertain, ontologies. This works was presented as a demonstration in the international conference CIKM 2017.
Based on this research, we collaborated with Inria (France) to investigate the emerging field of computational fact checking in search for further applications. This led to a survey paper presented at the WWW 2017 companion events, and a tutorial presented at both WWW 2018 and VLDB 2018.
Finally, in collaboration with a researcher of Mannheim University (Germany), we investigated the use of recent machine learning approaches, namely graph embedding and factorization machines, to infer the validity time of facts in knowledge graphs. We first extended the state of the art on non-temporal knowledge graph to graphs with temporal annotations (such as Wikidata or YAGO3). Then we showed how non-temporal facts can be used as features to learn the validity of temporal facts. This led to the publication of a workshop paper. This work is being extended to use of neural networks, which allows in particular to works with multiple levels of time resolutions (centuries, year, days, etc).
|
Research Products
(4 results)
