| With the developing of web and information technology,information on Internet is increasing at an exaggerate rate.However,the accessible resource for user of all the information seems like only a drop in the ocean.In that case,personal recommendation technology comes up.Traditional recommendation system relies on the information from a single domain.Nevertheless,with the development of Internet,more and more platforms are connected to each other,the resource in a single domain could not accommodate users’ need any longer.Moreover,the traditional single domain recommendation system is facing with other problems such as data sparsity,cold start problem,etc.Therefore,cross-domain recommendation will change this dilemma by incorporating multiple domain information and promoting the accuracy and diversity of personal recommendation technology.The advantages of cross-domain recommendation are making use of the data from multiple fields to model the user rating patterns and offering users the items from different domain to improve the diversity of prediction,which has become a research topic in industry and academic domain.Common recommendation systems for both single-domain or cross-domain case are mainly based on users’ rating data and in most cases recommendation is reduced to rating prediction,which leads to the restriction of rating sparsity.Therefore,more variety of data is taken into consideration to improve the performance of recommendation such as social tags.Tags are relevant keywords which can help people to describe or classify the resources more easily.It is believed that tags may reflect the user preference or item attribute.Nowadays,a large amount of websites adopt tags to improve the users’ satisfaction and it will be the foundation for tag-based recommendation system.In a nutshell,the innovation of this paper is combining the tag information with the framework of cross-domain recommendation.The proposed model offers the new avenues of methods to use tag information.Finally,we validated the effectiveness of our proposed models on real world dataset. |
