Machines’ ability to learn the behavior of complex real world systems has been the main research focus in temporal knowledge graphs (TKG). However, combining the human’s input - as part of a real-world TKG - into the modeling process has not yet been investigated. To fill this gap, we propose a novel human-centric machine learning (HCML) framework for TKG link prediction. The main goal is to demonstrate the value of a human-machine online optimization coupled with the self-attention mechanism. We argue that the joint development of a human-machine TKG model can detect low-signal information about the evolution of the graph that can have a significant impact on the dynamics. Finally, our proposed HCML framework is discussed on the basis of the European alternative fuels market as an exemplary use case with the outlook of the approach.