基于用户的协同过滤算法的改进研究

Improvement and Research of User-based Collaborative Filtering Algorithm

本文针对传统的基于用户协同过滤算法的稀疏性大,推荐效率,精度低等问题,提出了一种改进的算法。在计算"用户-评分"矩阵时,通过建立"项目-用户"倒查表,忽略无相同评分项的用户间相似性的计算,降低了用户评分矩阵的稀疏性,以及传统方法中对所有用户两两计算相似度的工作量。在计算用户相似度时,考虑到项目热门程度对推荐结果的影响,通过"惩罚"用户共同兴趣列表中的热门项目,避免了传统算法中由于赋予所有项目相同权重给个性化推荐结果带来的负面影响。最后,通过和数据集检验该算法,并且用十折交叉方法验证结果。结果表明,改进后的算法节约了运行时间,提高了推荐算法的效率和个性化。

Aiming at data sparseness of matrix, low efficiency and precision of recommendation existing in tra- ditional user-based collaborative filter in g algorithm, an improvement recommendation algorithm is put forward in this paper. The algorithm introduces item-user inversion table and calculates users＇ sim ilarity of whom with the same rating items in user- rating-data matrix. Therefore, it can reduce data sparseness o f matrix and avoid huge workload, which is brought by the calculation of pair-wise users＇ sim ilarity in traditional algorithm. Considering the impact of different hot degree of items in users1 sim ilarity calculation, the improvement algorithm punishes the negative in flu-ence of popular projects on users＇ common interest lists. Therefore, it gets around negative impact of popular pro-jects in the results of personalized recommendation. The experiment results of 10-fo ld cross-validation in M oviel- enslOOK and Het Rec 2011 open dataset shows that improved algorithm could save running time, increase efficiency and personalization of recommendation.