<p>Click - through rate forecasting represents an essential component in online advertising, and capturing interactive features is the key to the research. Existing methods grounded in deep learning cannot capture high - order and low - order interactive features in a controllable manner at both the bit dimension and the vector dimension. In this paper, we propose a combined bit-wise and vector-wise interactive features network for CTR prediction. In order to obtain low-order and high-order interactive features at these two levels, our model adopts a parallel structure. For bit-wise feature interactions, our model captures interactive features based on logistic regression(LR) and deep neural network(DNN) respectively. For vector-wise feature interactions, we have proposed a multi-head vector-wise network with residual connections to capture interactive features. The outcomes of the experiments validate the efficacy of the model for two public datasets compared with other state-of-the-art CTR prediction models (the AUC value increased by 0.35% and logloss decreased by 0.24% for Criteo of the proposed model).</p>