The complexity and depth of Information Extraction becomes increasingly apparent as time goes on. Heuristics, shocastic and more recently, neural models have proved challenging to scale into and out of various domains. In this paper we discuss the limitations of current approaches and explore if transferring human knowledge into a neural language model could improve performance in an deep learning setting. We approach this by constructing gazetteers from existing public resources. We demonstrate that leveraging existing knowledge we can increase performance and train such networks faster. We argue a case for further research into leveraging pre-existing domain knowledge and engineering resources to train neural models.
|Title of host publication||Artificial Intelligence XXXIV - 37th SGAI International Conference on Artificial Intelligence, AI 2017, Proceedings|
|Editors||Miltos Petridis, Max Bramer|
|Number of pages||6|
|Publication status||Published - Dec 2017|
|Event||37th SGAI International Conference on Artificial Intelligence, AI 2017 - Cambridge, United Kingdom|
Duration: 12 Dec 2017 → 14 Dec 2017
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||37th SGAI International Conference on Artificial Intelligence, AI 2017|
|Period||12/12/2017 → 14/12/2017|
Bibliographical noteFunding Information:
1 Implementation: https://github.com/zhiweiuu/SGAITagger. 2 This work is partially supported by the EPSRC (Grant REF: EP/P031668/1).
© Springer International Publishing AG 2017.
Copyright 2017 Elsevier B.V., All rights reserved.
- Information extraction
- Named entity recognition
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)