Research And Optimization Of FolkRank Tag Recommendation Algorithm

In recent years,the Internet has entered a stage of rapid development,at the same time,a variety of websites have also been rapidly developed,label system has slowly become the Web2.0 era at home and abroad research scholars are very interested in the hot direction.Tags can help users classify and annotate items.More and more users use tags to annotate items.On the one hand,the label can express the user's interest;On the other hand,the label also represents the description information of the item.An excellent label system can often give the user a satisfactory list of label recommendations when the user labels the item,so as to improve the quality of the user to label,and the label recommendation algorithm is generated.The common tag recommendation algorithms include collaborative filtering recommendation algorithm,content-based recommendation algorithm and graph-based recommendation algorithm.The graph-based tag recommendation algorithm FolkRank can effectively utilize the relationship between users,items and tags to achieve better tag recommendation performance.However,FolkRank algorithm does not take into account the internal relationships among user-user,item-item and tag-tag.In addition,the length of recommendation list of FolkRank and other top-n tag recommendation algorithms is usually fixed,which will lead to the decline of recommendation accuracy and poor user experience.In view of the above two problems,the innovation of thesis is as follows:1.In view of the problem that the FolkRank tag recommendation algorithm does not make full use of the user-user and item-item internal connections in the figure,thesis proposes an improved FolkRank tag recommendation algorithm,which can more fully consider the user-user and item-item internal relationships in the tag system.The above work is mainly completed from two aspects:(1)find the neighboring items of the target item according to the attribute information,and obtain the neighboring users of the target user according to the user's historical behavior of marking items.(2)calculate the weighting degree of items,evaluate the importance of items,and give initial weight to neighboring items according to the importance;The initial weight is assigned to the neighboring user based on the behavior of the user marking the item.2.Aiming at the problem that the FolkRank tag recommendation algorithm recommends the list length to be a fixed value,which leads to the decrease of the recommendation accuracy,thesis proposes an algorithm to optimize the top-n recommendation list length of the tag recommendation algorithm.The above work is mainly completed in two aspects:(1)first,add the tags greater than 1/2 of the top-1 tag score to the list of candidate recommendations.By defining the confidence index of paired tags,calculate the correlation between the tags in the list of candidates and the top-1 tag.(2)for the reordered tag recommendation list,by calculating the correlation coefficient of each sublist,the sublist with the highest correlation coefficient is the optimal recommended list3.length.In thesis,experiments were carried out on multiple data sets and compared with common algorithms such as FolkRank.The results show that the above two methods have good recommendation performance.