一种利用属性重心剖分模型的时间调整协作过滤推荐算法

Collaborative Filtering Recommendation Algorithm with Time Adjusting Using Attribute Center of Gravity Model

针对传统协作过滤推荐算法在相似性度量环节所存在的不足之处,提出一种利用属性重心剖分模型的时间调整协作过滤推荐算法,通过对项目属性矩阵填充用户所在群体所对应的普遍评分值,进而对用户—项目评分矩阵填充评分预测值,再在填充后的用户—项目评分矩阵的基础上,利用属性重心剖分模型度量出初步相似性,并结合传统相似性,得出复合相似性,最后对复合相似性进行时间维度调整,得到最终的相似性.仿真实验结果表明,与传统的协作过滤推荐算法相比,该算法可以获得更高的推荐精准度,并能够很好地适应于数据集极度稀疏、冷启动、用户兴趣漂移等特殊情形.

Faced with the shortcomings of traditional collaborative filtering algorithms＇ similarity measurement part, a collaborative filtering recommendation algorithm with time adjusting using attribute center of gravity model is proposed. First,the algorithm fills average ratings of the user＇s user group to the item-attribute rating matrix, and then fills rating predictions to the user-item rating matrix. Second, the algorithm uses attribute center of gravity model to calculate the preliminary similarity based on the filled user-item rating matrix, and then combines with the traditional similarity to gain the composite similarity. Last, the algorithm makes time-dimensional adjustment to the composite similarity to get the final similarity. Experiment results show that the algorithm can gain higher recommen- dation accuracy compared with the traditional collaborative filtering recommendation algorithms, and can well adapt to special circum- stances, such as data set＇s extreme sparsity, cold start, user interest drift, and so on.