A Parallel Deep Neural Network Using Reviews and Item Metadata for Cross-Domain Recommendation

Hong, Wenxing; Zheng, Ning; Xiong, Ziang; Hu, Zhiqiang

doi:10.1109/access.2020.2977123

Cited by 21 publications

(12 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Root-Mean-Square Error (RMSE) measures the average magnitude error that follows the quadratic scoring rule. The RMSE is defined as the square root of the average squared differences between the actual and prediction observations The RMSE is expressed as shown in the equation 13 (13) Where, is the actual value is the predicted value is the number of observations…”

Section: Resultsmentioning

confidence: 99%

“…The objectives of the model are to identify the user's point of interest, recommending products/services based on the user's latent interests [11]. In this situation, recommender systems have emerged as an effective mechanism to provide personalized recommendation services, which can effectively alleviate the information overload problem [12,13].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Accelerated - Generic Gradient Descent for E-Commerce Recommender Systems

Suresh¹,

Belinda²

2021

INDJCSE

View full text Add to dashboard Cite

The online product reviews play an important role in the field of e-commerce as rating of the product infers the preference for customers and they rely on them while making purchases. The main aim of providing an efficient recommender system is to gain useful information for user based on quality, worth and quantity of the product. The reviews given by the customers are used to extract the features of products that evaluate the similarity content reviewed by another customer. Therefore, the reviews written by the distinct persons would be similar, but with different words of expression shows difficulty during review differentiation. The developed model performed mining customer review for a set of manufactured goods and their characteristics are used as product features. The proposed Gated Recurrent Unit with Accelerated -Generic Gradient Descent (GRU-AGGD) technique extracts the relevant and sufficient features from the review to overcome the existing optimization problems. These relevant features are the summary of the product used to predict the rank using the users-attributes and users-items matrix and these two matrixes give rise to Non-Negative Matrix Factorization model to provide recommendations to user for the product. The experimental results of product recommender system are evaluated using the proposed GRU-AGGD that shows the improvement in terms of performance measures. The average MSE and RMSE values obtained for the proposed GRU-AGGD technique are 0.84 and 0.91 respectively, which are higher when compared to the Bayesian recommender models.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Accelerated - Generic Gradient Descent for E-Commerce Recommender Systems

Suresh¹,

Belinda²

2021

INDJCSE

View full text Add to dashboard Cite

show abstract

“…In addition, the authors introduced an attention mechanism to select useful reviews, incorporating representations of ratings as query vectors. Hong et al [20] developed a cross-domain recommendation model based on deep neural network. The model used a joint learning strategy to learn features of users and items from both target domain and source domains.…”

Section: Related Workmentioning

confidence: 99%

Research on BP Neural Network Recommendation Model Fusing User Reviews and Ratings

Wang

Hong

2021

IEEE Access

View full text Add to dashboard Cite

Sparsity of rating data is a severe problem to be solved in modern recommendation researches. The fusion recommendation method is an effective solution to the problem. The method combines rating data and other types of user feedback data, such as reviews and image to improve performance of the traditional recommendation algorithms. Some researchers have proposed fusion recommendation algorithms based on BP (Back Propagation) neural network and achieved some results. However, some existing fusion recommendation algorithms based on BP neural network still have some shortcomings. They rely on the assistance of the traditional recommendation algorithms. And the high complexity of the fusion process of these algorithms would have a negative impact on the fusion effect. In this paper, we modify the fusion recommendation algorithm and propose the NNFR (neural networks fusion recommendation) model. This model improves the structure of BP neural network by specially designing the structure of network layers. User reviews and ratings can be processed in two separate sub-networks respectively and further fused in the fusion layer. The fusion features of user reviews and ratings are directly applied to perform recommendation to avoid the assistance of the traditional recommendation algorithms and improve the fusing efficiency and quality. Experimental results show that the NNFR model can achieve better predictions than comparative recommendation algorithms to generate more accurate topk recommendation lists for users. Moreover, NNFR model can still produce high-quality recommendation results in the scenarios of sparse data.

show abstract

“…The review texts given by users on items in another domain are said to contain rich information, which is mined and transferred to the target domain [18,19]. As an improvement, [20] combined the mined review texts with item metadata in the source domain for recommendation accuracy in the target domain. Machine learning is another approach that has been widely utilized to improve recommendations in the target domain in CDRS [21].…”

Section: A Cross-domain Recommender Systemsmentioning

confidence: 99%

Privacy-Preserving Matrix Factorization for Cross-Domain Recommendation

2021

View full text Add to dashboard Cite

Cross-domain recommender systems are known to provide solutions to the cold start and data sparsity problems in recommender systems. This can be achieved by leveraging sufficient ratings and users' profiles in one domain to enhance accurate recommendations in another domain. However, domains with sufficient ratings are not willing to share their users' ratings with other recommender systems or domains due to users' privacy and legal concern. Hence this shows a need for a privacy-preserving mechanism that encourages secure knowledge transfer between different domains. This study proposes a privacy-preserving cross-domain recommender system based on matrix factorization. Specifically, the study formally described the privacy requirements of a cross-domain recommender system, which are different from a single domain recommender system. It designs a new framework for a privacy-preserving cross-domain recommender system and then utilized the somewhat homomorphic encryption (SWHE) scheme to ensure users' privacy. The SWHE scheme was used to encrypt users' ratings in different domains, shared latent factor approach was implemented between the domains and extracted knowledge was securely transferred from the source domain to the target domain. We prove that users' privacy is secured throughout the stages involved in the proposed protocol. Experiments on both synthetic and real datasets demonstrate the efficiency of our protocol.

show abstract

A Parallel Deep Neural Network Using Reviews and Item Metadata for Cross-Domain Recommendation

Cited by 21 publications

References 38 publications

Accelerated - Generic Gradient Descent for E-Commerce Recommender Systems

Accelerated - Generic Gradient Descent for E-Commerce Recommender Systems

Research on BP Neural Network Recommendation Model Fusing User Reviews and Ratings

Privacy-Preserving Matrix Factorization for Cross-Domain Recommendation

Contact Info

Product

Resources

About