We introduce partially separable temporal exponential random graph models for dynamic signed weighted networks. This extension enables the joint modelling of edge weight dynamics and sign changes over time, capturing both the intensity and the sentiment of relationships in evolving networks. By allowing partial separation between temporal dynamics and structural dependencies, our approach preserves interpretability while accommodating complex interactions between network structure, edge strength, and relational polarity. This advancement provides a flexible and expressive tool for analysing the evolution of polarisation strength across social, political, and economic networks. To examine the performance and reliability of our proposed model, we conduct a simulation study under a realistic controlled scenario.