Abstract
The emergence of the recommendation system has effectively alleviated the information overload problem. However, traditional recommendation systems either ignore the rich attribute information of users and items, such as the user’s social-demographic features, the item’s content features, etc., facing the sparsity problem, or adopt the fully connected network to concatenate the attribute information, ignoring the interaction between the attribute information. In this paper, we propose the information fusion-based deep neural attentive matrix factorization (IFDNAMF) recommendation model, which introduces the attribute information and adopts the element-wise product between the different information domains to learn the cross-features when conducting information fusion. In addition, the attention mechanism is utilized to distinguish the importance of different cross-features on prediction results. In addition, the IFDNAMF adopts the deep neural network to learn the high-order interaction between users and items. Meanwhile, we conduct extensive experiments on two datasets: MovieLens and Book-crossing, and demonstrate the feasibility and effectiveness of the model.
Highlights
We first propose a new recommendation model, the information fusion-based deep neural attentive matrix factorization (IFDNAMF) recommendation model, in which we introduce the auxiliary information to assist the model in describing the user features and item features more comprehensively and ; we propose a new method of information fusion, in which the inner product between the different information domains is adopted to learn the cross-features, Algorithms 2021, 14, 281
To make full use of the auxiliary information to model user preference and item representation more comprehensively and accurately, and, at the same time, learn the complex non-linear high-order interaction between users and items, and further improve the recommendation performance, this paper proposes information fusion-based deep neural attentive matrix factorization recommendation model, in which, when fusing information, element-wise product operation is employed to complete the interaction between features of different information domains in a targeted manner
Model with the multi-layer hidden layer network, referred to as generalized matrix factorization (GMF)+MLP, which is utilized to demonstrate the effectiveness of deep neural networks to model high-order interactions between the user and the item; Concat: The concat is the variant of the IFDNAMF model, which is the common method for traditional recommendation models to process the attribute information
Summary
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. During the inner product of traditional matrix factorization, the results of all dimensions are accumulated with the same weight, which could be seen as the connection weights all being 1, to get the final scalar result Both of these make it non-effective to model the complex non-linear relationships between the user and the item. We first propose a new recommendation model, the information fusion-based deep neural attentive matrix factorization (IFDNAMF) recommendation model, in which we introduce the auxiliary information to assist the model in describing the user features and item features more comprehensively and ; we propose a new method of information fusion, in which the inner product between the different information domains is adopted to learn the cross-features, Algorithms 2021, 14, 281.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
