Knowledge graph embedding (KGE) techniques learn node embedding from a KG and that is concatenated with the original word representation.

San Jose State University

Knowledge graph is a structured way of representing facts in the form of connections between entities ,relations and semantic descriptions. Different methods to study the relationship between KG and input/output sequences are

- Incorporate Knowledge Graph Embeddings into Language Generation
- Transfer Knowledge into Language Model with Knowledge Triplet Information.
- Perform Reasoning over Knowledge Graph via Path Finding Strategies.
- Improve the Graph Embeddings with Graph Neural Networks


NLG Enhanced by Knowledge Graph 

A Survey of Knowledge-Enhanced Text Generation

- Commonsense reasoning
- Dialogue system
- Creative writing


 NLG applications enhanced by knowledge graph

Incorporate Knowledge Graph Embeddings into Language Generation

Inconsistency in the vector space of entity embeddings and word embeddings paved the way for fine tuning language models directly on knowledge graph triplets.

Transfer Knowledge into Language Model with Knowledge Triplet Information

Path-based methods explore various patterns of connections among entity nodes ,they learn from walkable paths on KGs to provide auxiliary guidance for the generation process.There are two types of path ranking methods 

- Path routing and ranking
- Reinforcement learning based path finding


Perform Reasoning over Knowledge Graph via Path Finding Strategies

Graph-based representations aim at aggregating the context/neighboring information on graph data. They have been applied to both encoding and decoding phases.

Learn Before

Related