Abstract: Recently study shows that traditional pseudo-relevance feedback may bring topic drift．Therefore,to avoid topic drift effectively,it is essential to identify relevant documents and to form the pseudo relevant documents to user’s query．In this paper,based on clustering XML search results,a method was proposed to find good feedback documents．Firstly,a cluster-label extraction method based on equalizing weights was introduced,by fully considering the content and structure features in XML documents．Secondly,a two-stage ranking strategy was presented,as the candidate cluster ranking model and document ranking model．Finally,experimental data shows that compared to original retrieving method, the ranking models obtain better performance and find more relevant XML documents.

Key words: Information retrieval,XML pseudo-relevance feedback,XML search results clustering,Cluster label,Ran-king model