Positive relevance feedback algorithm based on dynamic weight query point movement
WU Yue-shu · Jisuanji gongcheng yu sheji · 2009
In order to overcome the problem of various weighted constants uncertainty in the query point movement algorithm, and the phenomenon that decaying vector constituted by the positive feedback samples is non-optimization in the weighted query point movement method, a new positive relevance feedback algorithm is presented. When determining the optimal query vector, the algorithm took the former memorized optimal query vector as the decaying vector, and used the feedback samples which the user selected to determine the decaying factor dynamically. Finally, the decaying vector and the user selected feedback samples continuously updated the optimal query vector, and accumulated more user retrieval intention information for the follow-up feedback. Comparing to other algorithms, experimental results show that this algorithm can accelerate the optimal vector adjusting speed and improve the retrieval performance.