Learning feature interactions is crucial for model performance in online recommendations. Extensive studies are devoted to designing effective structures for learning interactive information in an explicit way and tangible progress has been made. However, the core interaction calculations of these models are artificially specified, such as inner product, outer product and self-attention, which results in high dependence on domain knowledge. Hence model effect is bounded by both restriction of human experience and the finiteness of candidate operations. In this paper, we propose a generalized interaction paradigm to lift the limitation, where operations adopted by existing models can be regarded as its special form. Based on this paradigm, we design a novel model to adaptively explore and optimize the operation itself according to data, named generalized interaction network(GIN). We proved that GIN is a generalized form of a wide range of state-of-the-art models, which means GIN can automatically search for the best operation among these models as well as a broader underlying architecture space. Finally, an architecture adaptation method is introduced to further boost the performance of GIN by discriminating important interactions. Thereby, architecture and operation adaptive network(AOANet) is presented. Experiment results on two large scale datasets show the superiority of our model. AOANet has been deployed to industrial production. In a 7-day A/B test, the click-through rate increased by 10.94%, which represents considerable business benefits.

Recently, network embedding has been successfully used in recommendation systems. Researchers have made efforts to utilize additional auxiliary information (e.g., social relations of users) to improve performance. However, such auxiliary information lacks compatibility for all recommendation scenarios, thus it is difficult to apply in some industrial scenarios where generality is required. Moreover, the heterogeneous nature between users and items aggravates the difficulty in network information fusion. Many works tried to transform user-item heterogeneous network to two homogeneous graphs (i.e., user-user and item-item), and then fuse information separately. This may limit the representation power of learned embedding due to ignoring the adjacent relationship in the original graph. In addition, the sparsity of user-item interactions is an urgent problem need to be solved. To solve the above problems, we propose a universal and effective framework named Gemini, which only relies on the common interaction logs, avoiding the dependence on auxiliary information and ensuring a better generality. For the purpose of keeping original adjacent relationship, Gemini transforms the original user-item heterogeneous graph into two semi homogeneous graphs from the perspective of users and items respectively. The transformed graphs consist of two types of nodes: network nodes coming from homogeneous nodes and attribute nodes coming from heterogeneous node. Then, the node representation is learned in a homogeneous way, with considering edge embedding at the same time. Simultaneously, the interaction sparsity problem is solved to some extent as the transformed graphs contain the original second-order neighbors. For training efficiently, we also propose an iterative training algorithm to reduce computational complexity. Experimental results on the five datasets and online A/B tests in recommendations of DiDiChuXing show that Gemini outperforms state-of-the-art algorithms.