Design And Development Of The Collective Wisdom Of The Label Recommendation System

The21st century is not only the era of rapid development of the network, but the era for users to create wisdom. Network provides us with an unlimited of resources to choose, to share, but it also brings the problem of overload information. In recent years, the fast development of the personalized recommendation technology in the area of electronic commerce is a good solution to solve this problem. The biggest feature of Web2.0era is "user-centric". As the core value of Web2.0era, the Collective Wisdom has highlighted the importance of users. Social tagging system is not only the typical type of collective wisdom in the network environment, but one of the iconic application of Web2.0era. Therefore tags has brought new opportunities for personalized resource recommendation, which reflects the user’s preference for resources, but also the keyword description of resource characteristics.This paper mainly studies how to use tag’s information in social tagging system to execute personalized recommendation for users. This paper mainly has completed the following five work items:(1) The paper introduced the background of the topic, highlighted the necessity and urgency to design and develop the Tag Recommendation System based on the Collective Wisdom. Meanwhile, it analyzed the achievements and problems of the Collective Wisdom and Tag Recommendation System domestic and international, and designed feasible implementations to these issues.(2) It described related theories of Collective Wisdom, Tagging Systems, Tag Recommendation System, as well as the characteristics of collaborative filtering technology, content-based recommendation technology and hybrid recommendation technology.(3) It designed four kinds of tag recommendation algorithm based on the requirements of tag recommendation system function:Popular Tags (the most popular tag recommendation algorithm in a system),Item Popular Tags (the most popular tag recommendation algorithm in a resource), User Popular Tags (the tag recommendation algorithm that users often use) and TagBaseSIM (improved tag recommendation algorithm based on content). The paper mainly introduced TagBaseSIM algorithm which was improved on the traditional content-based tag recommendation algorithm. First, it geted the users’preference information by using the user model and resource model; And then put all the characteristics tagged by target users in the resource and the tags in similar resources together as a candidate tags set; Third, getting the similarity in each cluster and the resources in clustering analysising on the candidate label sets;Finally,the list of tags from the cluster are recommended to the given user based on their values of similarity. Due to the charcterristic of tags,they may have the problems to induce the recommendation accuracy,such as the ambiguity or redundancy of Tag,so using the method of K-means tag clustering can reduce the impact on the recommendation quality problems effectively.(4) On the data mining platform of Sql Server2005and Weka, we use offline experimental methods on the real data sets in social bookmark system Delicious to do ten-fold cross-validation experiments ten times. First, do experiments on the former three algorithm in the data set to obtain the accuracy of each algorithm, the recall rate and the comprehensive evaluation F1; Then compared the improved content-based tag recommendation algorithm proposed in this paper and the traditional content-based tag recommendation algorithm. The results show that the improved content-based recommendation algorithm has better recommendation accuracy.(5) Finally,we design and develop a tag recommendation system whose theme is "growth and development in IT teachers" on the platform of MyEclipse combined with MySql database management system.The system uses four kinds of tag recommendation algorithms in different pages to recommend a tag list to the users, which overcomes the cold-start problem.