Survey on Embedding Models for Knowledge Graph and its Applications
Manita Pote
TL;DR
This survey surveys knowledge graph concepts, large-scaleKG resources, and deep learning approaches for KG embedding, contrasting translation-based methods with neural network–based models and highlighting their training objectives such as margin-based losses for $h + r \approx t$ and related variants. It covers architectures (TransE, TransR, DistMult, ComplEx, NTN, R-GCN) and neural models (SME, ConvKB, KBGAN), illustrating how embeddings enable NLP tasks like link prediction and triple classification. The review then details four application threads—fake news detection, drug-related analytics, suicidal ideation detection, and enriching KG with social-media data—demonstrating practical impact in information integrity, health-related insights, and knowledge augmentation. Overall, KG embeddings provide compact, semantically rich representations that improve reasoning, data integration, and downstream NLP tasks, with promising directions in misinformation analysis and propaganda detection.
Abstract
Knowledge Graph (KG) is a graph based data structure to represent facts of the world where nodes represent real world entities or abstract concept and edges represent relation between the entities. Graph as representation for knowledge has several drawbacks like data sparsity, computational complexity and manual feature engineering. Knowledge Graph embedding tackles the drawback by representing entities and relation in low dimensional vector space by capturing the semantic relation between them. There are different KG embedding models. Here, we discuss translation based and neural network based embedding models which differ based on semantic property, scoring function and architecture they use. Further, we discuss application of KG in some domains that use deep learning models and leverage social media data.