An intelligent recommendation approach for online advertising based on hybrid deep neural network and parallel computing
- PDF / 1,762,086 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 85 Downloads / 274 Views
(0123456789().,-volV)(0123456789(). ,- volV)
An intelligent recommendation approach for online advertising based on hybrid deep neural network and parallel computing Zilong Jiang1 • Shu Gao1 Received: 25 April 2019 / Revised: 31 May 2019 / Accepted: 16 July 2019 Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract Artificial neural network is an information process system that simulates the structure and intelligent behavior of the brain based on the understanding of its organizational structure and operating mechanism. In order to deal with the recommendation task for online advertising, this paper proposes a special intelligent recommendation approach based on hybrid deep neural network. This approach integrating embedding mapping network, factorization machine, stacked denoising autoencoder and regression model, can effectively model complex categorical data, learn higher-order abstract features like the brain, and then classify them to achieve the purpose of precise recommendation, and can be well parallelized. Experimental results on the real dataset show that compared with the baseline models, the proposed approach has better performance in the face of scenarios containing a large number of complex categorical data. Keywords Hybrid deep neural network Intelligent recommendation Brain-inspired design Parallel processing
1 Introduction Computation intelligence (Intelligent algorithm) is an effective computing tool to address specific problems by imitating some aspects of intelligence of natural organisms through computer programs. For example, artificial neural network (ANN) is a brain-inspired information processing method, which uses multiple neuron models to imprecise simulate the structure of the brain. As an improvement of traditional neural network, deep neural network has become the mainstream method of artificial intelligence research at present, and has achieved fruitful research results in image processing, natural language processing, intelligent recommendation and other fields. In order to deal with the advertising recommendation task in complex scenario, based on the advantages of deep neural network, we proposes an intelligent recommendation approach called FMSDAELR in this study.
& Zilong Jiang [email protected] 1
The main advantages and contributions can be summarized as follows: • The proposed approach makes the most of embedding mapping network and factorization machine that can efficiently model high-dimensional sparse input data, and obtains the low-dimensional preliminary representation of input data, which contains the information of each feature and the low-order feature interactions. • Good ability of extracting higher-order abstract features. Denoising autoencoder (DAE) simulates the human brain’s ability to draw correct conclusions based on incomplete information to some extent. The proposed approach uses multiple DAEs to construct a deep neural network with the function of distributed information storage and parallel cooperative processing, which has
Data Loading...
