On the effectiveness of evaluating retrieval systems in the absence of relevance judgments