隐含因子在随机游走模型中的应用

Application of latent factor in random walk model

研究了随机游走模型中转移概率矩阵的计算方法,并针对推荐系统中的随机游走模型,提出了一种基于隐含因子的方法来计算转移概率矩阵。这种方法通过矩阵分解的技术将随机游走模型中的节点映射为隐含因子向量,并使用该因子向量计算节点之间的转移概率。为了验证上述方法,针对具体的随机游走模型进行了实验,结果表明该方法不仅具有很好的稳定性,而且在数据集极端稀疏的情况下,也能有效地捕捉节点之间的转移概率,从而提高随机游走模型的推荐性能。

This paper studied the calculation methods of the transition probability matrix in random walk models. It proposed a method based on latent factors to obtain the transitive probability matrix of random walk models for the recommendation system. The method mapped the nodes in the random walk model to latent factor vectors through matrix factorization and computed the transitive probabilities between nodes using the vectors. A specific random walk model demonstrated the validity of the method. The experimental results show that the method has good stabilities and can improve the prediction performance of random walk model by effectively capturing the transitive probabilities between nodes especially when the data is extremely sparse.