Introduction to Graph Machine Learning/

...

/

Translation-Based Methods

Translation-Based Methods

Learn about translation-based knowledge graph embedding methods.

We'll cover the following...

Translation

In translation-based models, we use the distance-based function to generate the embeddings. There are multiple ways to compute the distance functions in Euclidean space.

Some of the translation-based algorithms are listed below.

TransE

The idea behind TransE is that it represents the entities as a vector in the high-dimensional space and represents the relation as a translation vector.

Press + to interact
TransE
TransE

If we consider a triple (h,r,t)(h,r,t), the TransE algorithm tries to find embeddings for all entities and relations so that the combination of head and relation vectors results in the vector of the tail, i.e., h+r≈th + r ≈ t.

The TransE algorithm learns the embeddings of entities and relations by minimizing the difference between the predicted vectors and the actual vectors. For this, the scoring function (ff) can be formulated as the distance (L1 or L2 norm) between h+rh + r and tt.

fr(h,t)=−∣∣h+r−t∣∣f_r (h, t) = −||h + r − t|| ...