K-means 算法在隐语义模型中的应用磁

摘要

Latent Factor Model(LFM ) is an important model widely used in text mining .It has the advantage of high precision and low memory cost in rating prediction .However LFM model is not suitable for processing large‐scale sparse ma‐trix .In order to improve the performance ,K‐means algorithm is introduced to deal with rating data into LFM .This new model is called K‐LFM .First of all ,K‐means is used to classify user and item information in K‐LFM .And then the rating matrices are refactored to reduce the scale and sparse degree of orignal matrix .Finally training model with refactoring matix , can get predict rating .The experiment on public data set movielens shows that K‐LFM model is superior to LFM model on processing efficiency .Besides ,the prediction accuracy isn't significantly affected .%隐语义模型（LFM ）是文本挖掘领域的重要模型，将它应用于推荐系统的评分预测具有预测精度高和占用内存小的优点。但由于时间开销较大，LFM 模型并不适合用于处理大规模稀疏矩阵。针对此问题，论文将 K‐means 算法引入到 LFM 模型的评分数据处理，得到改进模型 K‐LFM 。在 K‐LFM 模型中，利用 K‐means 算法对评分矩阵中的用户和项目数据进行聚类处理，然后重构评分矩阵降低原始矩阵的稀疏程度和矩阵规模，最后用重构后的评分矩阵训练模型，预测评分。通过在 movielens 数据集上实验发现 K‐LFM 模型在运行时间上较 LFM 模型有大幅降低，而预测精度没有受到明显影响。