Abstract: At present, research on knowledge graph-based question answering is primarily focused on the English domain. In contrast, the development in the Chinese domain lags behind due to its late start and the lack of sufficient question and answering (Q&A) corpus. The existence of numerous aliases for entities in Chinese further increases the complexity of question answering. To address these challenges, a knowledge graph-based question answering model PBJ (Prompt-Biaf-JointPN) has been proposed, in which biaffine model and pointer networks are jointly integrated. First, the Prompt module adopts a chain-of-thought approach to decompose multi-hop questions into multiple single-hop questions. Then, for each single-hop question, the Biaf module employs a biaffine model as the output layer to perform entity mention recognition, while the JointPN module unifies entity disambiguation and relation matching as a multiple-choice reading-comprehension task to obtain intermediate answers. Next, through a recursive iterative mechanism orchestrated by the Prompt module, the answer from the previous hop is fed as the entity mention for the next-hop question until the final complete answer is generated. To evaluate the model's performance, experiments were conducted on three datasets (NLPCC-ICCPOL, KgCLUE, and NLPCC-MH) against eleven existing models. The results demonstrate the PBJ model's accuracy: its BJ variant (Prompt module removed) exceeds the top single-hop baseline Pipeline2 w/Biaf by 2.15% in Hits@1, while the full PBJ surpasses the leading multi-hop baseline ChineseBERT by 1.26%, confirming its capability for complex question answering in the Chinese context. Finally, ablation studies were conducted to assess the individual contributions of the proposed Prompt, Biaf, and JointPN components. The results reveal that each module is instrumental to PBJ's perfor-mance: removing Prompt, Biaf, or JointPN respectively degrades Hits@1 by 2.13%, 2.84%, and 6.51% relative to the full model. Consequently, by jointly modeling entity disambiguation and relation matching, the PBJ model effectively mitigates the bottlenecks of abundant entity aliases and scarce corpora in Chinese KGQA, offering a practical and scalable new avenue for Chinese multi-hop question answering.