Distmult

EMBEDDING ENTITIES AND RELATIONS FOR LEARNING AND INFERENCE IN KNOWLEDGE BASES

ICLR 2015

年代久远，虽然文章里是这样的，但是现在大家都按照直接相乘来计算。Mult

abstract

1. most exsiting model under a unfied framework, where entities are low-dimensional vectors learned from a neural network and relations are bilinear and/or linear mapping funtions.

2. evaluate different choices of entity representations and relation representations under this framework on canonical link prediction task.
   show a simple bilinear and/or linear formulation achieve new SOTA.

3. a novel approach that utilizes the learned embeddings to mine logical rules.

bilinear-diag: DistMult
TransE: DistADD
文章里说是： h*M_r*t， 但是网上查的教程都说：
好像就是改了个得分函数，h+t-r改成了h*t*r。。。
不知道为什么变成直接乘了。。
看了pykg2vec的代码，也不知道rel怎么限制成diagonal的。。。

TA-DistMult

将三元组加入时间，将relation改成： relation+时间状语(since, until..) + year + month + day 这样的一个序列p

将p过一个lstm，

最后当成一个三元组h*p*t就可以了。。