基于标签和协同过滤的改进推荐算法研究

针对基于标签和协同过滤的个性化推荐(TCF)没有考虑评分数据的作用和用户兴趣标签稀疏的问题,提出了一种加入评分数据并扩展用户兴趣标签的基于标签和协同过滤的改进推荐算法(ITCF).首先,以项目- 标签相关度构造项目特征向量,并结合评分构造用户特征向量和用户-标签关联度; 其次,对用户的历史偏好标签集进行基于标签相似性和基于近邻用户偏好的扩展; 最后,以MovieLens数据集为例对ITCF算法的有效性进行实验验证.实验结果表明,在稠密的数据集中,ITCF算法的平均准确率和平均召回率比文献[2]和[3]算法的平均准确率和平均召回率分别提升约2.0%和1.7%; 在稀疏的数据集中,当推荐项目数不超过20时,ITCF算法的平均准确率和平均召回率约比文献[2]和文献[3]算法的平均准确率和平均召回率分别提升约0.2%和0.8%.因此,本文提出的ITCF算法具有较好的应用前景.

Research on improved recommender algorithm based tag and collaborative filtering

Aiming at the problem that the personalized recommendation based on tag and collaborative filtering(TCF)does not consider the function of rating data and spare user interest tags, this paper proposes an improved recommendation algorithm based on the tag and the collaborative filtering(ITCF), which adds rating data and extends user interest tag. Firstly, the item feature vector is constructed by the item-tag correlation degree, and the user feature vector and the user -tag correlation degree are constructed by combining the rate. Secondly, the tag set of users' historical preference is extended based on tag similarity and nearest neighbor user preference. Finally, the experiment is carried out to verify the efficiency of ITCF algorithm on the MovieLens data set. The experimental results show that compared with that of reference [2] and [3] TCF algorithm, the precision and recall rate of the proposed ITCF algorithm are increased by 2.0% and 1.7% in the dense data set, and 0.2% and 0.8% in the sparse data set when the number of recommended items does not exceed 20. Therefore, the ITCF algorithm proposed in this paper has a good application prospect.