UTD's Event Nugget Detection and Coreference System at KBP 2017

Jing Lu and Vincent Ng.
Proceedings of the 2017 Text Analysis Conference, 2017.

Click here for the PDF version.

Abstract

We describe UTD's participating system in the event nugget detection and coreference task at TAC-KBP 2017. We designed and implemented a pipeline system that consists of three components: event nugget identification and subtyping, REALIS value identification, and event coreference resolution. We proposed using an ensemble of 1-nearest-neighbor clasifiers for event nugget identification and subtyping, a 1-nearest-neighbor classifier for REALIS value identification, and a learning-based multi-pass sieve approach consisting of 1-nearest-neighbor classifiers for event coreference resolution. Though conceptually simple, our system compares favorably with other participating systems, achieving F1 scores of 50.37, 40.91, and 33.87 on these three tasks respectively on the English datset, and F1 socres of 46.76, 35.19, and 28.01 on the Chinese dataset. In particular, it ranked first on Chinese event nugget coreference.

BibTeX entry

@InProceedings{Lu+Ng:17b,
  author = {Jing Lu and Vincent Ng},
  title = {UTD's Event Nugget Detection and Coreference System at KBP 2017},
  booktitle = {Proceedings of 2017 Text Analysis Conference},
  year = 2017}