Prototypical networks relation classification model based on entity convolution. (January 2023)

• Record Type: Journal Article
• Authors: Feng, Jianzhou; Wei, Qikai; Cui, Jinman
• Is Part Of: Computer speech & language. Volume 77(2023)
• Publication Date: 2023-01
• Subjects: Relation classification -- Few-shot learning -- Entity convolution -- Prototypical networks; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2022.101432
• Languages: English
• ISSNs: 0885-2308

Abstract

Abstract: Sentence-level relation classification is a technique for classifying the relation between the head entity and the tail entity in a sentence. Currently, it is popularly used to realize relation classification based on deep learning methods. However, these methods rely heavily on large-scale annotated data, and the role of head and tail entities' information is not fully explored. In response to the above problems, we propose a prototypical networks model based on entity convolution for relation classification, which deforms the head entity and tail entity vectors encoded by BERT into multiple different convolution kernels and then performs convolution operations on the original sentence. Thus we can extract the features related to the entities in the sentence and classify the extracted features by using prototypical networks to realize relation classification. Experimental results strongly demonstrate that our model achieves state-of-the-art results compared with baseline models. Highlights: We utilize a few-shot learning method based on the prototypical network to implement relation classification, thereby alleviating the insufficient data in the new domain. We propose a new method for entity convolution to alleviate the problem about the under-utilization of text information. We evaluate our method on FewRel 1.0 and 2.0 datasets, and experimental results demonstrate that our model achieves state-of-the-art results compared with baseline models.