摘要:
针对物品推荐技术中存在的目标用户和相似用户均未标记的物品无法预测以及未考虑用户已购买情况等问题，提出了一种基于图像内容的物品推荐方法。该方法提取物品图像的颜色、形状和纹理特征来表示物品，通过图像内容间的相似性和用户相似性完成了对目标用户未标记物品兴趣度的预测，并引入用户兴趣度因子来反映用户已购买的情况对用户兴趣的影响，最终根据用户购买项目的静态特征给出推荐结果。在自建的物品图像数据集上，用文中方法与基于用户的协同过滤技术、基于项目的协同过滤技术以及two-way三种方法进行对比试验，实验结果表明，该方法具有良好的物品推荐品质。
Abstract: The technology used to recommend products suffers from the problems such as inability to recommend products unrated by neither target user nor his similar users and ignoring the previous consumptions of users. To address the problems mentioned above, a new recommendation method based on image content is proposed. This recommendation method describes the product by the color, shape and textual of product images, and is able to recommend new products and the products unrated by target user and his similar users by considering the similarities of images and users, and reflects the impact on user’s interest by user’s consumption. Finally, the system gives the recommendation result based on the static feature of the items bought by users. We test our algorithm on an image dataset built by ourselves. The experimental results show that the new method has better capacity for recommendation compared to user-based collaborative filtering, item-based collaborative filtering and two-way method.