Applying Data Mining to Pseudo-Relevance Feedback for High Performance Text Retrieval