Predicting new relational facts with the help of very few samples.

San Jose State University

KGC adds new triples to the existing knowledge graph and also accomplishes subtasks like link prediction, entity prediction and relation prediction. KGC also learns low dimensional embedding for triple prediction with recent advancement on capturing multi step relationship paths. KGC also do triple classification which evaluates the correctness of factual triple.

 - Embedding-based Models:
 - Relation Path Reasoning
 - RL-based Path Finding
 - Rule-based Reasoning
 - Meta Relational Learning:
 - Triple Classification


Knowledge Graph Completion 

A Survey on Knowledge Graphs: Representation, Acquisition and Applications

Embedding-based ranking methods learn embeddings from existing triples and replaces  head and tail entities with all available entities and scores are calculated for each triple formed and top k triples are ranked.

Embedding-based Models

This method learns the complex relation path information.

Relation Path Reasoning

Deep reinforcement learning is used for finding out multiple hop reasoning paths.

RL-based Path Finding

KGC can be used for logical rule learning.

 Rule-based Reasoning

Meta Relational Learning

Triple classification is a binary classification to test whether the triples are actually correct.

Learn Before

Related