Personalized Search Based On Social Annotation

As the development of search engine, users have more requirement for information retrieval. To better satisfy it, personalized search method is invented. The main propose of these methods are solve the problem that how to give results that based on user’s interest when in retrieval. In this process, more and more resource is add to it to build users’ interest. Because of the development of Web 2.0, social annotation system is well popularize. Social annotations are the resource which come from users directly, so it’s very useful for personalized search. By the time, using social annotations in personalized search is tested and is proved to be effective. But there are still some way to advance it. The main purpose of this paper is to explore how to use social annotations to raise the effect of personalized search.There are something which should be pay attention to when use social annotations. One of them is that annotations are one side of users’ interest, which can’t completely represent it. And the annotations from different users may have different quality, which can influence the result. Besides, the sparsity of annotations is also difficult to solve. Depend on these reason, the main contribution of this paper are:First, when only use social annotations, we promote a personalized search method based on the network of users’ similarity. We use a method which is similar to VSM to calculate the weight of users’ tags, and we add the similarity of users on the same document to it. Then we consider the relation of similar users as a network, and use the method like the interaction of nodes to solve the sparsity problem and reduce complexity of search. So that the result will be more reliable. The experiment prove that the method can improve the effect and reduce the consumption.Then for better constructing user interest, we promote a personalized search method reconcile with user similarity and user quality. We add web categories to the method that only use annotations to calculate users’ similarity. And we also user these two resource in the modified Social PageRank to judge users’ quality, so that the extension of annotation and the quality of retrieval will be improved. The experiment shows that this method can obviously improve the result.