Question Answering (QA) has become a hot topic since the proposal of chatGPT. The QA system based on Knowledge Graph (KG)s or Temporal Knowledge Graph (TKG)s relies heavily on the graph structure and link prediction. To improve the performance of link prediction, this paper focuses on the investigation of ensemble learning techniques. Since deep learning-based methods can dynamically assign weights for each model and provide more accurate weights than other methods, it is investigated in this paper to predict weights as the hyperparameters of ensemble models, specific to TKGs. Two neural network architectures are proposed to predict ensemble weights and evaluated based on six metrics. To analyze the results, it is found that the neural network-based ensemble model not only outperforms all the individual models but also the grid search-based ensemble model. ATiSE has the best performance across all base models. However, after the grid search-based ensemble, the ensemble model performed 33% better than ATiSE in HITS@1. Moreover, after the neural-network-based ensemble, the ensemble model performed 8% better than the grid search-based ensemble model in HITS@1. Moreover, it is worth noticing that in the case that only predicts one element of a fact (predicting only head/relation/tail/timestamp), the neural network-based ensemble model improves better when predicting tail and timestamp elements.