Research Of Personalized Tag Recommendation Technology Based On The Graph Model

With the rapid development of Internet technology, tagging system has beenwidely used and become an important part of many Web2.0websites. Tagging systemis an application which provide function of tagging to user, it become very popularbecause of its convenient operation and easy to use. Delicious, Flickr, Last.fm, andCiteULike are the typical tagging systems. Tag recommendation is an importantfunction of the tagging system, which provide suitable tags to the user and these tags aregenerally interested user or comply with the characteristics of the user. Tagrecommendation both user-friendly operating and improve the quality of tagging.User, tag and item form a three-dimensional mode in the tagging system. In orderto use the traditional method, existing tag recommended method mostly processedthree-dimensional model by splitting into multiple binary relations, and use a simpleway to express the information of tag. This conversion process prone to loss ofsemantics, which lead to the personalized information of tag has not been utilized. taghas great significance to both user and item, which should be fully considered. Both theactive users and users who has less tags in the tagging system. The cold start usershould be considered when making personalized tag recommended.For the existing methods has some insufficient, this paper presents a personalizedtag recommended method based on graph model. This method considers the integrityof user, tag and item, transforming the three-dimensional mode to a weighted tripartitegraph. For the adjacent vertices, we use an integrated measure of the weight, whichconsider both the popularity and the personalized features of the tag. For thenon-adjacent vertices, we obtained the weight by the thought of shortest path. Finally,we get the list of personalized tags by considering the relationship between user and tag,tag and item.We evaluate our method on two real-world folksonomies collected from CiteULikeand Last.fm. The experimental results demonstrate that the proposed method improvesthe recommendation performance and is effective for both active taggers and cold-starttaggers compared to existing algorithms.