A Personalized Re-ranking Algorithm Based on Relevance Feedback

Relevance feedback is the most popular query reformulation strategy. However, clicking data as user's feedback is not so reliable since the quality of a ranked result will influence the user's feedback. An evaluation method called QR (quality of a ranked result) is proposed in this paper to tell how good a ranked result is. Then use the quality of current ranked result to predict the relevance of different feedbacks. In this way, better feedback document will play a more important role in the process of re-ranking. Experiments show that the QR measure is in direct proportion to DCG measure while QR needs no manual label. And the new re-ranking algorithm (QR-linear) outperforms the other two baseline algorithms especially when the number of feedback is large.