Ongoing challenges in knowledge graphs and GraphRAG
KGs are a powerful medium for storing and organizing information, but there are still limitations and open questions. Especially for large KGs, scalability is important; a balance must be struck between expressiveness and computational efficiency. Plus building a KG requires a lot of computational effort (using an LLM to extract triplets from a large corpus of text can be expensive and require adequate infrastructure). In addition, once the KG is built, it must be evaluated and cleaned, which also requires some effort (manual or computational). Moreover, growth in the KG also means growth in the infrastructural cost to enable access or use. Querying large KGs requires having optimized algorithms to avoid the risk of increasingly large latency times. Industrial KGs can contain billions of entities and relationships, representing an intricate and complex scale. Many of the algorithms are designed for small-scale KGs (up to thousands...
