In order to address the problems of low relevance and lack of diversity of end-to-end models in generating responses in current open-domain dialogue systems, a Multi-Source Knowledge-Enhanced Dialogue Generation Framework (MSGF) is used to investigate various aspects of dialogue generation in this study. Firstly, multiple different knowledge sources are integrated to improve the knowledge coverage related to dialogue background information. Secondly, the adoption of the global knowledge selection module can avoid topic conflicts between different knowledge sources, thereby eliminating the problem of confusion in the meaning of dialogue topics. In addition, the model also introduces a fusion prediction module to generate responses by obtaining information from different knowledge sources. The results show that the MSGF model outperforms other similar models with more comprehensive knowledge coverage and higher topic relevance of generated responses. It is concluded that the proposed MSGF model is capable of understanding the content of dialogue conversations and significantly improving the performance of dialogue systems.