Research On Cross-domain Recommendation Method Based On Transfer Learning

In the era of big data,Internet users are producing data,creating data,and using data every day.The amount of data is growing exponentially.The problem of information overload is becoming more and more serious.At the same time,users are unable to find their own information in a timely and effective manner in the face of a large amount of information.Information required.In order to solve this problem,the recommendation system came into being and has been widely used in various fields,but traditional recommendation is still concentrated in a single field.With the rapid development of Internet technology,the number of users and items has increased dramatically,making the traditional single recommendation system Facing the severe challenge of data sparseness and cold start.Transfer learning helps the target domain to complete the learning task by transferring the knowledge-rich source domain data to the target domain with sparse knowledge,while effectively alleviating the data sparsity and cold start problems in the target domain.Based on the idea of transfer learning,combined with the recommendation algorithm,a cross-domain recommendation method has emerged,which aims to recommend the target field by integrating knowledge from different fields,thereby improving the accuracy of the target field recommendation and increasing the diversity of recommendations.At present,cross-domain recommendation has become a hot topic in the recommendation system,but most of the existing cross-domain recommendation methods based on transfer learning are mostly based on a single scoring mode between domains for migration,without much consideration of the inter-domain scoring scale Differences and domain relevance issues,resulting in unsatisfactory migration effects,negative migration issues,and reduced recommendation accuracy.In response to these problems,the main research work of this article is as follows:(1)Combine the user’s knowledge of scoring items and the user’s own behavior knowledge,and assist the target field in task learning together.At present,the more common user behavior knowledge,such as the labels given by the user to the item and the user’s own attribute tags,these tags can not only indicate the implicit characteristics of the item,but also reflect the user’s own behavior preferences.Considering the difference in the score scale between different domains and whether the domains are related,combined with label features,a cross-domain recommendation method based on label feature migration CR-BLFM(Cross-domain recommendation method based on la-bel feature migration)is proposed.Using tags as a bridge for knowledge transfer between different domains,firstly,the user item rating matrix of the source domain is decomposed by the method of non-negative orthogonal matrix decomposition to reduce the dimension of the matrix,and then the source domain is clustered by the K-Means clustering method.Users cluster to get different types of user clusters;secondly,learn the labels used by different user clusters through a neural network,and train a classifier that can identify different types of users based on user labels;finally,based on the same type of users The similarity between the similarity and the weight of the known score to predict the unknown score.(2)If only the correlation between tags is considered and the semantic correlation between different tags is ignored,the effectiveness of recommendation will be reduced.Therefore,this paper proposes a cross-domain recommendation method CR-BLSM based on tag semantic migration(Cross-domain recommendation method based on label semantic migration),using label semantic information as a bridge for knowledge transfer between different domains,first of all,through word2vec technology to represent the potential characteristics of all labels between different domains,and then by clustering Form different tag clusters and use them as a common embedding space between the source and target domains;second,map the tag feature vectors of users and items in the source and target domains into this embedding space to identify similar users across domains And items;Finally,by transferring knowledge between similar users and items,the recommendation effect of the target domain is improved.(3)Combined with the label information shared in different fields,a cross-domain recommendation system between the book field and the film field was designed and implemented.The user can recommend the corresponding book for the user by logging into the system and selecting the corresponding label.When the user enters the book’s detail page,the user will recommend the corresponding type of movie to the user based on the book’s label and make a corresponding recommendation Explanation.