Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics 2019
DOI: 10.18653/v1/p19-1130
|View full text |Cite
|
Sign up to set email alerts
|

Exploiting Entity BIO Tag Embeddings and Multi-task Learning for Relation Extraction with Imbalanced Data

et al.

Abstract: In practical scenario, relation extraction needs to first identify entity pairs that have relation and then assign a correct relation class. However, the number of non-relation entity pairs in context (negative instances) usually far exceeds the others (positive instances), which negatively affects a model's performance. To mitigate this problem, we propose a multitask architecture which jointly trains a model to perform relation identification with crossentropy loss and relation classification with ranking lo… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
20
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
4
3
1

Relationship

0
8

Authors

Journals

citations
Cited by 33 publications
(20 citation statements)
references
References 28 publications
0
20
0
Order By: Relevance
“…In the experiments, we use two English benchmark datasets for RE, namely, ACE2005EN (ACE05) 5 and SemEval 2010 Task 8 (SemEval) 6 (Hendrickx et al, 2010). For ACE05, we use its English section and follow previous studies (Miwa and Bansal, 2016;Christopoulou et al, 2018;Ye et al, 2019) to pre-process it (two small subsets cts and un are removed) and split the documents into training, development, and test sets 7 . For SemEval, we use its official train/test split 8 .…”
Section: Datasetsmentioning
confidence: 99%
“…In the experiments, we use two English benchmark datasets for RE, namely, ACE2005EN (ACE05) 5 and SemEval 2010 Task 8 (SemEval) 6 (Hendrickx et al, 2010). For ACE05, we use its English section and follow previous studies (Miwa and Bansal, 2016;Christopoulou et al, 2018;Ye et al, 2019) to pre-process it (two small subsets cts and un are removed) and split the documents into training, development, and test sets 7 . For SemEval, we use its official train/test split 8 .…”
Section: Datasetsmentioning
confidence: 99%
“…Different tasks may use different types of loss functions. For example, in [141], the cross-entropy loss for the relation identification task and the ranking loss for the relation classification task are linearly combined, which performs better than single-task learning. Specifically, given 𝑀 tasks each associated with a loss function L 𝑖 and a weight 𝜆 𝑡 , the overall loss L is defined as…”
Section: Loss Constructionmentioning
confidence: 99%
“…The second configuration is designed to deal with data imbalance (cf. Table 1), following recent studies that show that jointly learning common characteristics shared across multiple tasks can have a strong impact on RE performances [34,29]. To this end, we jointly train two classifiers using multitask objectives.…”
Section: Data and Annotationmentioning
confidence: 99%