IJCAI填坑2 已看论文

填坑。。。

SoRec: Social Recommendation Using Probabilistic Matrix Factorization

想要解决的问题：

建模社交信任关系

Contribution:

(1) Our method can deal with missing value problem, while their methods cannot. (2) Our method is interpreted using a probabilistic factor analysis model.
(3) Complexity analysis shows that our method is more efficient than their methods and can be applied to very large datasets.

建模的方法：

Social Network Matrix Factorization

U是latent user matrices,Z是factor feature matrices

高斯和贝叶斯

然后还要考虑一个情况是：如果一个人相信大部分人，那么他的rating基本也没什么用；如果一个人被大部分人相信，那么他的rating可能用处更大，所以这里加了个权。

User Item Matrix Factorization

和社交那块建模类似

最终的优化目标和梯度：

实验

效果还不错吧

Conclusion

In the future, we will explore whether the distrust information is useful to increase the prediction quality, and how to incorporate it.

Reference

可以关注下
Maximum Margin Matrix Factorization:MMMF
Probabilistic Matrix Factorization: PMF
Constrained Probabilistic Matrix Factorization: CPMF