Abstract: The sparse user-item matrix often hurts the performance of recommendation system. Aiming at this problem, an object image recommendation algorithm based on multi-instance learning is proposed. The images are regarded as bags and the segmented regions as instances. The Diverse Density(DD) function is adopted to search the maximum DD point; Next the positive and negative images is used to measure the users’ similarity. The similarity result is integrated with the traditional cosine similarity. Experimental results show that, the algorithm can improve the recommendation performance.

Key words: object image recommendation, collaborative recommendation, multi-instance learning, diversity density function, combination recommendation

摘要： 用户评分矩阵稀疏问题影响协同过滤的推荐性能。为此，提出一种基于多示例学习的对象图像推荐算法。将分割区域的视觉特征作为图像中的示例，利用多样性密度函数求得最大多样性密度点，使用正负图像内容评价不同用户间的相似性，将其与传统余弦相似性进行组合，从而实现推荐。实验结果表明，该算法提高了推荐性能。

关键词: 对象图像推荐, 协同推荐, 多示例学习, 多样性密度函数, 组合推荐

CLC Number: 

• TP301.6