Research On Cross-Domain Personalized Recommendation Based On Tags

With the continuous development of network and information technology,the number of resources on the Internet presents an exponential growth trend.How to dig out the information which is consistent with people’s actual needs from the huge amount and different forms of information is always the key and difficult point in domestic and foreign research.Personalized recommendation technology is a kind of information search technology based on user needs and actual tasks.At present,many scholars at home and abroad have studied personalized recommendation methods and applied their results to different fields such as social networks and e-commerce.However,the current recommendation tasks often have problems such as insufficient personalized recommendation,cold start and sparse data,resulting in an unsatisfactory final recommendation effect.Cross-domain recommendation is a recommendation method that extracts relevant data from one or more other fields so as to supplement and expand the data of the target fields.It can better solve the problem of insufficient data in the current recommendation tasks.The premise of cross-domain recommendation is to explore the correlation between two or more domains.As a kind of metadata used to describe users and resources,social tags widely exist in different fields.Based on them,it is easy to mine the relationship between different fields to achieve cross-field recommendation.In this paper,DBSCAN algorithm,a clustering algorithm with high compatibility with social tags,was improved.On this basis,the characteristics of social tags in different fields were analyzed.Based on the improved clustering algorithm and tag characteristics analysis results,cross-domain personalized recommendation based on social tags was realized.In the improvement process of DBSCAN algorithm,this paper firstly selects and evaluates tags.On this basis,the tag and resource data of users are collected from douban.com,and the tag vector,resource vector and user vector were constructed.Based on the original DBSCAN algorithm,the tag clustering was carried out.Through the analysis of the clustering results,the existing clustering methods were found to have shortcomings,and the corresponding improvements were made in combination with the characteristics of social tagging.After the algorithm is improved,the improved algorithm is used as the basis for the tag clustering analysis,and the clustering effect is measured to verify the effectiveness of the improved algorithm.In the process of cross-domain recommendation based on tags,this paper firstly analyzes the tagging behavior of users in different fields,and controls the quality of tags used for cross-domain recommendation through the analysis of user behavior.On the basis of user behavior analysis and data quality control,the improved DBSCAN algorithm combined with TF-IDF and other data mining methods realizes the cross-domain recommendation based on content,collaborative filtering and mixed model.The cross-domain recommendation model constructed in this paper is based on the data of users in a single domain to realize cross-domain recommendation and realize the diversification of resource recommendation.When recommending resources,the recommendation of similar resources in different fields is realized,solving the problems of cold start and data sparsity.