Dual Denoising Autoencoder Based on Neighbor-Attention Module for Implicit Feedback Recommendation

In recent years, the content-aware recommendation of implicit feedback data using denoising autoencoder has become the mainstream technology in the field of recommendation systems. However, the technology still faces the following main problems: the denoising autoencoder structure ignores the intrinsic structural duality property of the model, and the recommendation of traditional implicit feedback data based on the denoising autoencoder ignores the neighbor item information. To solve these problems, a dual denoising autoencoder based on neighbor-attention module (DDAENAM) was proposed. The encoder and the decoder were designed as a dual closed loop, and the dual attribute of the structure was used to train the encoder and the decoder at the same time, such that the feedback signal between the encoder and the decoder can be shared in the model. Then, a neighbor-attention module was proposed to extract word embedding and neighboring item information. Results show that in three standard implicit feedback datasets including CiteULike-a, MovieLens-20M, and Amazon-Books, DDAENAM achieves the best result compared with the state-of-the-art models, with an average improvement of 2.4%. Under the review of evaluation indices select recall rate, DDAENAM is slightly lower than the joint representation learning model (JRLM), except for the MovieLens-20M. The proposed DDAENAM in this study achieves the best result with the combination of three modules including attention module, term– attention module, and neighbor–attention module, indicating the effectiveness of the DDAENAM module © 2021 School of Science, IHU. All rights reserved