The Ongoing Affair of Knowledge Graphs and LLMs
Check — Knowledge Graph Series [https://ogre51.medium.com/knowledge-graph-series-01aa961c0d21]
Knowledge Graphs excel in capturing relationships. They also have greater reasoning abilities. But they are difficult and costly to construct.
On the other hand, LLMs contain extensive knowledge. But they are prone to bias, hallucinations and other issues. They are also computationally expensive to fine-tune or to adapt for specific domains.
To harness the benefits of both methods, knowledge graphs and LLMs can be integrated together in several ways.
Knowledge Graphs represent data in a network/graph. They represent concepts, events and entities as nodes, and the relationships between them as edges. These relationships (edges) determine the context of the node, and consequently, allow for better understanding of a word’s semantics and distinguishing between its multiple possible meanings.
Now, both Knowledge Graphs and LLMs can be queried to retrieve knowledge.
Knowledge Graphs locate answers by finding relevant connected nodes. And LLMs are prompted to fill in the [MASK] token to complete the…