Research On Method Of Evaluating Confidence Of Social Annotations

The main feature of Web2.0is allowing users to create content. As an application ofWeb2.0environment, social annotation system has become the most popular applicationdue to its lower accidence, more flexible operation, and easier to use. Social annotationtechnology is helpful to improve the quality of information retrieval and seek newinteresting information from mass web resources. Social annotation can reflect thesociality of tags, it is a manifestation of public wisdom. However, social annotationbelongs to forksonomy, the quality of tags (referred to as tag confidence) is irregularowning to arbitrary and incorrect annotating, which decreases the effect of tags to organize,share, and seek information.In order to make social annotation to serve applications information retrieval relatedbetter, we propose a confidence evaluating model of social annotation. Firstly, we analyzeand calculate the influence factors: the influence of annotated users, the semanticsimilarity between documents, and the semantic similarity between tags. The frequency ofidentical annotating behavior between users can be seen as the influence of users; Inaccordance with vector space model, documents can be expressed as keyword vectors, wecan regard the distance between vectors as the semantic similarity between documents;Tag can be represented as a combination vector of user information and resourceinformation, the distance between tag vectors can be thought as the semantic similaritybetween tags. On these bases, we make out a model by covering all influence factorsabove to calculate the confidence of tags.We propose three methods to validate the accuracy of confidence model proposed: tagranking, text classification based on tags, and text clustering based on tags. Comparedwith other three tag ranking methods, the tag ranking method based on tag confidence isthe most approximate to manual ranking result. After we import the confidence evaluating model, the effect of text classification and clustering can achieve the best. The results ofour experiments demonstrate that the confidence evaluating model of social annotation isaccurate and can improve the application effect based on tags.