Research On Cross-domain Recommendation Of Goods Based On Dense Convolutional Network And Attention Mechanism

In recent years,the Internet has shown a highly resilient and dynamic development trend,leading to an exponential growth of online data resources,leading to serious "information overload" problems,and thus promoting the development of recommendation systems.However,data sparsity and user cold start issues remain the main challenges that constrain the improvement of traditional single domain recommendation performance.Cross domain recommendation,as one of the main ways to solve this problem,has received widespread attention.The main idea of cross domain recommendation is to utilize the rich information in the auxiliary domain to improve the recommendation effect in the target domain.Existing methods mostly start from inter domain user and product rating data,or only extract coarse-grained features from comments.The mining depth of the rich user preference information contained in it is not enough,making it difficult to fully utilize the existing data.In addition,different user scenarios can also have an impact on knowledge transfer between domains,resulting in a decrease in recommendation performance.Based on the above issues,this paper builds two cross-domain recommendation models under different user scenarios from the perspectives of user sharing and user non-sharing,based on dense convolutional network and attention mechanism.On the one hand,in the scenario where there are common users between domains,a cross-domain recommendation model for goods sharing by users between domains is proposed.After vectorization of related data,dense convolutional network is used to extract features from the scoring and commentary data in the field,which can effectively realize the reuse of features.Select a attention mechanism to fuse user and commodity ratings and text reviews within each field.In order to achieve score prediction in the target domain,the model uses BP network to achieve cross-domain mapping of relevant features,and finally builds user preferences in the target domain to achieve score prediction for users in the target domain.On the other hand,in response to the reality of having different customer groups in different fields,a cross domain product recommendation model considering non shared users between domains is proposed based on the former model.The model uses K-means++clustering to obtain domain personality knowledge by extracting user and commodity characteristics in various fields using dense convolutional network and attention mechanism.Identify overlapping parts of personality knowledge as common knowledge among domains,and migrate them to target domains with personality knowledge to achieve score prediction.Finally,experiments were conducted on the proposed model in the dataset in scenarios of inter domain user sharing and user non sharing.The results showed that compared to the comparative model,the effectiveness of the proposed model was significantly improved.High quality recommendations can also be made in the case of inter domain user non sharing,thus verifying the effectiveness of the model.This research work provides a solution to alleviate the problem of data sparsity and cold start.While providing users with more accurate product recommendations,it can effectively improve customer satisfaction of e-commerce platforms and improve the platform’s profit level.