【摘要】 针对数据稀疏性导致推荐算法准确度不高的难题,提出一种融合词频-逆向文件频率(Term frequency-inverse document frequency, TF-IDF)的受限玻尔兹曼机(Restricted Boltzmann machine, RBM)推荐算法,利用受限玻尔兹曼机构建用户项目二维评分矩阵,利用余弦相似度计算方法得出初始推荐评分,最后融合词频-逆向文件频率算法生成最终推荐结果集。对MovieLens1M的电影评分数据进行实验,结果显示,该文提出的混合推荐算法的平均绝对误差(Mean absolute error, MAE)和均方根误差(Root mean square error, RMSE)分别为0.602 8和0.622 5,比传统受限玻尔兹曼机分别提高3.22%与6.06%,也优于对照混合推荐模型的准确率。该算法能提高用户评分预测精度,进一步提升推荐质量。更多还原

【Abstract】 Aiming at the problem of low accuracy of recommendation algorithm caused by data sparsity, the accuracy of recommendation algorithm is improved.This paper proposes a score prediction algorithm based on restricted Boltzmann machine(RBM)and term frequency-inverse document frequency(TF-IDF).The restricted Boltzmann machine is used to fill in the score matrix to alleviate the problem of data sparsity.The cosine similarity calculation method is used to get the initial recommendation score. The TF-IDF algorithm is fused to generate the final recommendation result set. Experiments on MovieLens1 M dataset show that the MAE and RMSE of the hybrid recommendation algorithm are 0.602 8 and 0.622 5 respectively, which are 3.22% and 6.06% higher than those of the traditional restricted Boltzmann machine. This algorithm is also better than the highest accuracy of the control hybrid recommendation model. This algorithm can improve the accuracy of user rating prediction and further improve the quality of recommendation.更多还原