一类BP神经网络优化评分预测的协同过滤推荐算法

A Class of Collaborative Filtering Recommendation Algorithm with the Score Prediction Optimized by BP Neural Network

协同过滤推荐算法是当前被广泛关注的推荐算法之一,算法中基于Resnick公式的评分模块只考虑了用户的浏览行为,忽略了用户间的评分影响,而且在数据稀疏的情况下不能较好的发挥评分预测的作用。针对这一不足,提出了一类BP神经网络优化评分预测的协同过滤推荐算法。其中,BP神经网络优化评分预测模块首先将相似用户的评分频数作为输入数据,并将目标用户正确评分作为输出数据进行神经网络训练,然后使用训练完的神经网络对用户评分进行预测。应用该模块分别替换皮尔逊推荐、模糊混合用户推荐和改进后的Top-N推荐代表性算法中的Resnick公式,对应给出了3种BP神经网络优化评分预测的协同过滤推荐算法。Movie-Lens数据集上的试验结果表明,该类协同过滤推荐算法在个性评分预测、特别是在稀疏数据评分预测方面有较强的竞争力。

The collaborative filtering recommendation algorithm(CFRA)is one of the widely concerned recommendation algorithms in nowadays.The module based on Resnick formula in CFRA only considers users'browsing behavior,while ignores the interaction between users'scores.The Resnick formula can't play agood role in the case of data sparsity.To solve this problem,this paper proposes a class of collaborative filtering recommendation algorithms with the score prediction optimized by BP neural network.In the proposed algorithms,the module predicting the score by BP neural network takes the frequencies of the similar users'scores as input data,and the target users'score is used as the correct training data for neural network,then,the trained network is used to predict the scores of target users.Furthermore,the module predicting the score by BP neural network is applied to replace the Resnick formula in three representative algorithms,that is,the Pearson recommendation algorithm,Fuzzy-genetic approach based on novel hybrid user recommendation algorithm,and the improved Top-N recommendation algorithm.Thus,three collaborative filtering recommendation algorithms with the score prediction optimized by BP neural network are exhibited.The experiment results on Movie-Lens data set indicate the competitiveness of this kind of collaborative filtering recommendation algorithms for the personality score prediction,and the score prediction in the case of data sparsity.