This paper introduces a novel framework, Learning to Plan from Knowledge Graphs (LPKG), to enhance the planning capabilities of large language models (LLMs) in complex question-answering (QA) tasks. The framework leverages knowledge graph (KG) patterns to construct planning data, which is then used to fine-tune LLMs. The fine-tuned LLMs are better equipped to handle complex QA tasks involving retrieval. The authors also develop a comprehensive logical QA benchmark, CLQA-Wiki, to evaluate the performance of LLMs on complex QA tasks. Experiments on multiple datasets, including the newly proposed CLQA-Wiki benchmark, demonstrate the effectiveness of the LPKG framework and the benefits of KG-derived planning data. The results show that LPKG outperforms baseline methods on conventional complex QA datasets and achieves significant improvements on the CLQA-Wiki benchmark, highlighting the importance of using KG-sourced planning data.This paper introduces a novel framework, Learning to Plan from Knowledge Graphs (LPKG), to enhance the planning capabilities of large language models (LLMs) in complex question-answering (QA) tasks. The framework leverages knowledge graph (KG) patterns to construct planning data, which is then used to fine-tune LLMs. The fine-tuned LLMs are better equipped to handle complex QA tasks involving retrieval. The authors also develop a comprehensive logical QA benchmark, CLQA-Wiki, to evaluate the performance of LLMs on complex QA tasks. Experiments on multiple datasets, including the newly proposed CLQA-Wiki benchmark, demonstrate the effectiveness of the LPKG framework and the benefits of KG-derived planning data. The results show that LPKG outperforms baseline methods on conventional complex QA datasets and achieves significant improvements on the CLQA-Wiki benchmark, highlighting the importance of using KG-sourced planning data.