In the era of big data intelligence, knowledge has become the fundamental driving force for technological development. Big data knowledge fusion, which constructs a unified knowledge system by associating multi-source fragmented data, serves as a key support for enhancing knowledge completeness and improving cognitive intelligence. Based on basic paradigms and classic methods, the research status of three key links—knowledge representation learning, knowledge alignment matching, and knowledge conflict resolution—was systematically summarized. Meanwhile, new progress in the context of big data was deeply explored, covering frontier directions such as temporal knowledge fusion, cross-modal knowledge fusion, and large language model knowledge fusion. Furthermore, future development trends were prospected, and potential research issues, including the fusion of symbolic and parametric knowledge as well as cross-modal temporal knowledge fusion, were highlighted. Through a panoramic review of basic links and emerging paradigms, important reference and guidance are provided for theoretical expansion and technological evolution in the field of big data knowledge fusion.