Abstract: A method for constructing user preference profile in personalized image retrieval is proposed.This method is based on short-term interest and long-term interest，in which the short-tem interest vector is decided by analyzing the image low-lever features，and the long-term interest vector is deduced by collecting interests from inference engine.Experiments results show that the recall/precision are significantly improved and satisfied by personalized user as well.

Key words: user preference profile, personalized image retrieval, inference engine, relevance feedback

摘要： 针对个性化图像检索的语义鸿沟问题，提出了一种新的用户兴趣模型的构建方法。将用户兴趣模型分为长期兴趣和短期兴趣：用户的短期兴趣由图像的低层特征映射得到；用户的长期兴趣经过推理机推理，将短期兴趣映射为高层语义得到，从而弥补语义鸿沟。实验结果表明，经过用户兴趣模型过滤的图像检索结果符合用户的个性化要求，相比已有方法在查准率和查全率上取得了明显的改善。

关键词: 用户兴趣模型, 个性化图像检索, 推理机, 相关反馈