Knowledge Guided Metric Learning for Few-Shot Text Classification

Sui, Dianbo; Chen, Yubo; Mao, Binjie; Qiu, Delai; Liu, Kang; Zhao, Jun

doi:10.18653/v1/2021.naacl-main.261

Cited by 14 publications

(11 citation statements)

References 17 publications

(24 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Geng et al [62] extracted an RDFS schema (ontology) from NELL for augmenting zero-shot KG link prediction with unseen relations. Sui et al [165] extracted entity concepts (entity classes) from NELL for augmenting few-shot text classification, where entities are retrieved via exactly string matching using entity mentions in the text.…”

Section: Sub-kg Extractionmentioning

confidence: 99%

“…Input Mapping [85,114,125] Class Mapping [100] Joint Mapping [1,2,101,114,149,165,204,215,218,224] Data Augmentation…”

Section: Mapping-basedmentioning

confidence: 99%

“…As the method in [101], the class embeddings are not directly used in prediction, but used to constrain the learning of the sample mapping. Sui et al [165] proposed a KG-aware few-shot text classification method. It compares the testing sample with the few-shot samples of each unseen class using not only a task-agnostic relation network but also a task-relevant relation network which is able to apply diverse metrics for diverse tasks armed with external knowledge extracted from NELL.…”

Section: Input Mappingmentioning

confidence: 99%

“…Few-shot text classification is similar to zero-shot text classification: the majority of the solutions directly utilize different kinds of word embeddings while the research on KG-aware method is rare. Besides the ICD-9 augmented CNN classifier which was applied to both few-shot and zero-shot text classification [149], the joint mapping method recently proposed by Sui et al [165] utilizes knowledge retrieved from NELL for augmenting a network which calculates the matching of the input and the class. The existing text classification benchmarks can be adopted for evaluating KG-aware few-shot text classification.…”

Section: Text Classificationmentioning

confidence: 99%

“…Meta learning is usually applied by either reducing the parameter searching space in training using meta parameters such as more optimized initial parameter settings, or transforming a classification problem to a metric learning problem where a testing sample is matched with the unseen classes based on their few-shot samples and meta learned mappings. KGs have been utilized to optimise such meta learning-based methods; for example, Sui et al [165] retrieved relevant knowledge from a KG named NELL [124] to construct task-relevant relation networks as mapping functions for addressing few-shot text classification. Meanwhile, the aforementioned idea of knowledge transfer can also adopted for addressing FSL, where KG auxiliary information is becoming increasingly popular in recent years [35,139,174,219].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Zero-shot and Few-shot Learning with Knowledge Graphs: A Comprehensive Survey

Chen¹,

Geng²,

Chen³

et al. 2021

Preprint

View full text Add to dashboard Cite

Machine learning methods especially deep neural networks have achieved great success but many of them often rely on a number of labeled samples for training. In real-world applications, we often need to address sample shortage due to e.g., dynamic contexts with emerging prediction targets and costly sample annotation. Therefore, low-resource learning, which aims to learn robust prediction models with no enough resources (especially training samples), is now being widely investigated. Among all the low-resource learning studies, many prefer to utilize some auxiliary information in the form of Knowledge Graph (KG), which is becoming more and more popular for knowledge representation, to reduce the reliance on labeled samples. In this survey, we very comprehensively reviewed over 90 papers about KG-aware research for two major low-resource learning settings -zero-shot learning (ZSL) where new classes for prediction have never appeared in training, and few-shot learning (FSL) where new classes for prediction have only a small number of labeled samples that are available. We first introduced the KGs used in ZSL and FSL studies as well as the existing and potential KG construction solutions, and then systematically categorized and summarized KG-aware ZSL and FSL methods, dividing them into different paradigms such as the mapping-based, the data augmentation, the propagation-based and the optimization-based. We next presented different applications, including not only KG augmented prediction tasks in Computer Vision and Natural Language Processing (e.g., image classification, visual question answering, text classification and knowledge extraction), but also tasks for KG curation (e.g., inductive KG completion), and some typical evaluation resources for each task. We eventually discussed some challenges and future directions on aspects such as new learning and reasoning paradigms, and the construction of high quality KGs.

show abstract

Section: Sub-kg Extractionmentioning

confidence: 99%