Quantum Inspired Neural Network Model And Its Application In CTR Prediction

Click-through rate(CTR)prediction is an important research task in online advertising recommendation.It aims to predict the possibility of users clicking on advertisements in the advertising display.Recently,as neural networks have achieved good results in many fields,more and more researchers have proposed neural network-based models for CTR prediction tasks.Generally,these neural network-based CTR prediction models only learn the interaction of low-order and high-order features through the original features.However,the original features are usually high-dimensional sparse,which makes it difficult for the neural network model to learn effective feature interaction under a large number of parameters.Some scholars propose to use feature engineering to increase effective and dense features to improve model performance,but feature engineering requires a certain amount of domain knowledge and a large number of manpower,which is usually expensive and the introduction of feature engineering will make model migration in other fields or tasks difficult.Therefore,it is necessary to design a model that can automatically generate more dense features to solve this problem.Here,we propose a more efficient and accurate quantum heuristic neural network-based model for CTR prediction task.In this paper,a density matrix-based convolutional neural network module is designed to automatically extract dense and effective features,which take advantages of density matrix-based representation and convolutional neural networks.In this model,an input instance is constructed into a quantum system represented by a density matrix,and then the feature interaction in the density matrix is extracted using a convolutional neural network.After that,a CTR prediction model DMCNN(Density Matrix Based Convolutional Neural Network)is constructed.In this model,the original normalized features and the interaction features extracted from the density matrix based convolutional neural network model are concatenated and sent to a neural network for click rate prediction.The DMCNN model proposed in this paper has achieved good results on several well-known public data sets,which proves the effectiveness of this model.We combine the language expression advantages of quantum theory and traditional neural networks to the CTR prediction task,which provides a novel idea for the development of the CTR estimation task and has certain scientific significance.