LightPROF: A Lightweight AI Framework that Enables Small-Scale Language Models to Perform Complex Reasoning Over Knowledge Graphs (KGs) Using Structured Prompts

Abstraction: Lightweight Retrieve-Embed-Reason framework for small LLMs on KGs

Key points:

• LightPROF (Lightweight Prompt learning-ReasOning Framework) uses a Retrieve-Embed-Reason pipeline: relation-based retrieval, Transformer-based Knowledge Adapter embedding, and prompt-combined reasoning
• Achieves 83.7% accuracy on WebQSP and 59.3% on the harder CWQ dataset, outperforming state-of-the-art methods
• Compared to StructGPT: 30% reduction in processing time, 98% reduction in input token usage
• Plug-and-play: works with various open-source LLMs without full fine-tuning; only the Knowledge Adapter requires training
• Narrows retrieval scope by sampling KGs using stable relations as retrieval units rather than converting entire subgraphs to text
• Evaluated on Freebase-based datasets WebQuestionsSP (4,737 QA pairs) and ComplexWebQuestions (34,689 pairs)

Connections: Knowledge Graphs · Prompt Engineering · Large Language Models · Retrieval Augmented Generation

Source: https://www.marktechpost.com/2025/04/12/lightprof-a-lightweight-ai-framework-that-enables-small-scale-language-models-to-perform-complex-reasoning-over-knowledge-graphs-kgs-using-structured-prompts/