基于内存的协同过滤推荐系统存在数据稀疏和数据集异构的问题。为此,提出一种基于变权重相似度计算和自适应局部融合参数的协同过滤方法。通过统计数据集,提取用户一项目评分项的用户情感信息量计算用户相似度,同时根据用户一项目评分项的评分质量改进项目相似度计算方法,利用基于相似用户(或项目)的方法预测置信度,得到自适应局部融合参数,以增强协同过滤方法对数据集的适应能力。实验结果表明,相比传统全局融合参数方法,该方法在数据稀疏情况下的平均绝对误差降低了0.02,具有较高的推荐精度和推荐覆盖度,并且有效解决了数据稀疏和数据集异构问题。
Aiming at the problem of data sparsity and dataset heterogeneity in memory-based collaborative filtering recommendation system, this paper proposes a collaborative filtering method based on variable weight similarity computation and Adaptive Local Fusion- parameter(ALFP). The method extracts user emotion information of user-item rating by counting data set to compute user similarity, meanwhile, according to user-item rating quality to improve item similarity computation method. The method then gets ALFP to enhance collaborative filtering's adaptability to dataset by forecast confidence of user-based method and item-based method. Experimental results show that the method outperforms traditional Global Fusiou-parameter(GFP) method by 0.02 with Mean Absolute Error(MAE) in case of data sparsity, it has higher recommendation precision and recommendation coverage, and effectively solves the problem of data sparseness and heterogeneous data sets.