基于组合相似度的优化协同过滤算法

摘要

Collaborative filtering is a most widely used recommendation technique in personalised recommendation system.With the increase in numbers of user and item, the sparsity of data becomes an important factor affecting the recommendation quality.Therefore, the six-type combined similarities are presented, which combines two traditional similarity metrics of the adjusted cosine similarity and the Pearson correlation with the structure similarity metrics such as Jaccard coefficient, Salton coefficient and IUF coefficient.Experiment done on MovieLens shows that the combined similarity-based optimised collaborative filtering algorithm raises a lot in MAE, RMSE, recall, coverage and precision, and improves recommendation quality as well.%协同过滤算法是个性化推荐系统中应用最广泛的一种推荐技术。随着用户数量和项目数量的增加，数据的稀疏性成为影响推荐质量的重要因素。为此，将传统相似度指标修正余弦相似性、Pearson相似度，与结构相似度指标Jaccard系数、Salton系数、IUF系数进行组合，提出6种组合相似度。在MovieLens上的实验表明，基于组合相似度的优化协同过滤算法在平均绝对偏差MAE、均方根误差RMSE、召回率、覆盖率和确率等性能上都有了较大提高，提高了推荐质量。