Abstract: In view of the existing personalized recommendation system based on tags in the construction of user interest model which does not both fully mine user’s real interest and take into account the impact of the time factor on the recommendation results, an improved collaborative filtering algorithm based on tags（ITCF） is proposed. The algorithm integrates user’s ratings into calculation of the weight of tags, considers the impact of user’s different interest on the results of the project, and uses a combination of exponential gradual forgetting function and time window to capture the changes of user’s interest. Experiments on the Movielens data set show that the improved algorithm has better recommendation effects on three evaluation metrics which are precision, hit-rank and NDCG, the quality and effectiveness of the recommendation are both better than the traditional method.

Key words: collaborative filtering, tags, exponential gradually forgotten function, time window, precision

摘要： 针对现存的基于标签的个性化推荐系统在构建用户兴趣模型时未充分挖掘用户真正的兴趣爱好，且未考虑到时间因素对推荐结果的影响，提出一种改进的基于标签的协同过滤算法（ITCF）。该算法将用户评分融入到用户对标签权重的计算中，考虑用户不同兴趣程度的项目对推荐结果的影响，并使用指数渐进遗忘函数和时间窗口相结合的方法来捕捉用户兴趣的变化。在数据集Movielens上的实验证明，改进后的算法在precision、hit-rank以及NDCG三个评价指标上均取得了较好的推荐效果，其推荐的质量和效果均优于传统方案。

关键词: 协同过滤, 标签, 指数渐进遗忘函数, 时间窗口, 准确率