Inductive Learning on Commonsense Knowledge Graph Completion

Wang, Bin; Wang, Guangtao; Huang, Jing; You, Jiaxuan; Leskovec, Jure; Kuo, Chin-Hwa

doi:10.1109/ijcnn52387.2021.9534355

Cited by 21 publications

(24 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Text Feature Fusion [5,10,66,109,129,159,180,181,187,208,213,226,230] Multi-modal Feature Fusion [5,128,150,217] Table 1. A summary of KG-aware ZSL paradigms.…”

Section: Mapping-basedmentioning

confidence: 99%

“…The ontological information such as the entity's hierarchical classes are utilized by their word embeddings which are combined with the entity's word embeddings by concatenation, averaging or weighted averaging, and fed into a classification model. Wang et al [181] proposed a commonsense KG link prediction method named InductiveE which can deal with unseen entities by utilizing entity textual descriptions. It first represents an entity using the concatenation of its text embeddings by the fastText word embedding model [87] and the last layer [CLS] token of the pre-trained BERT [48], and then feed the entity representations of the graph into a model composed of an encoder -a gated-relational GCN and a decoder -a simplified version of ConvE [47] to predict each triple's score.…”

Section: Input Featuresmentioning

confidence: 99%

“…Since they represent a triple's entities and relation as features and feed these futures into a model for prediction, we also regard them as the category of text feature fusion. Different from [181] where BERT is used for initial but fixed entity representations, the As KG-BERT, Zha et al [213] also proposed to predict triples as a downstream text classification task of BERT, utilizing the text information of entities and relations. But they fine-tune BERT using not only single triples but also possible paths that connect two entities where reasoning is conducted explicitly.…”

Section: Feature Learning Modelsmentioning

confidence: 99%

“…To address unseen entities in zero-shot KG completion, the existing studies often utilize their auxiliary information such as names, textual descriptions and attributes, often using methods of the mapping-based paradigm [70,158] and the class feature paradigm [5,129,159,180,181,187,208,213,226]. With the explosive growth of zero-shot KG completion methods for unseen entities, various benchmarks have been proposed for evaluation.…”

Section: Kg Completion With Unseen Entitiesmentioning

confidence: 99%

See 3 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

“…Text Feature Fusion [5,10,66,109,129,159,180,181,187,208,213,226,230] Multi-modal Feature Fusion [5,128,150,217] Table 1. A summary of KG-aware ZSL paradigms.…”

Section: Mapping-basedmentioning

confidence: 99%

Section: Input Featuresmentioning

confidence: 99%

Section: Feature Learning Modelsmentioning

confidence: 99%

Section: Kg Completion With Unseen Entitiesmentioning

confidence: 99%

See 2 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

show abstract

“…WN18RR FB15K-237 YAGO3-10 Models MRR H@1 H@3 H@10 MRR H@1 H@3 H@10 MRR H@1 H@3 H@10 Malaviya et al (2020). Results for RotatE are from Wang et al (2020). Results for QuatE 2 are obtained using our implementation (QuatE 2 is a special case of BiQUE as discussed in Section 4.1).…”

Section: Evaluation Protocolmentioning

confidence: 99%

BiQUE: Biquaternionic Embeddings of Knowledge Graphs

Guo¹,

Kok²

2021

Preprint

View full text Add to dashboard Cite

Knowledge graph embeddings (KGEs) compactly encode multi-relational knowledge graphs (KGs). Existing KGE models rely on geometric operations to model relational patterns. Euclidean (circular) rotation is useful for modeling patterns such as symmetry, but cannot represent hierarchical semantics. In contrast, hyperbolic models are effective at modeling hierarchical relations, but do not perform as well on patterns on which circular rotation excels. It is crucial for KGE models to unify multiple geometric transformations so as to fully cover the multifarious relations in KGs. To do so, we propose BiQUE, a novel model that employs biquaternions to integrate multiple geometric transformations, viz., scaling, translation, Euclidean rotation, and hyperbolic rotation. BiQUE makes the best tradeoffs among geometric operators during training, picking the best one (or their best combination) for each relation. Experiments on five datasets show BiQUE's effectiveness.

show abstract