Learning Chain Of Thoughts Prompts for Predicting Entities, Relations, and even Literals on Knowledge Graphs

RALP (the project) represents a transition in the Knowledge Graph (KG) space from traditional embedding-based methods (KGE) to LLM-based prompting. Its technical core—using Bayesian Optimization via MIP to search for optimal CoT prompts—is sophisticated but fundamentally a specific optimization technique for a niche task. With 0 stars and 5 forks in its first 3 days, it is currently in the 'early academic release' phase. Its defensibility is low because the logic can be replicated or integrated into broader prompt optimization frameworks like Stanford's DSPy or Microsoft's GraphRAG. Frontier labs pose a high risk here; as LLM reasoning (like OpenAI's o1) becomes more natively capable of handling structured data and literals, the need for specialized prompt-learning algorithms for link prediction may vanish. The project’s value lies in its ability to handle 'literals' (text/numbers) which traditional KGEs struggle with, but this is a capability that general-purpose frontier models are rapidly absorbing. Its displacement horizon is 1-2 years as GraphRAG architectures and automated prompt engineering become standardized.