DistMultEmbedding<T>
Models & Types · AiDotNet.RetrievalAugmentedGeneration.Graph.Embeddings
DistMult embedding model: bilinear diagonal scoring with Σ(h_k · r_k · t_k).
Inheritance: KGEmbeddingBase<T> → DistMultEmbedding<T>
For Beginners
DistMult is the simplest bilinear model:
- Each entity and relation gets a vector
- Score = element-wise product of head, relation, and tail vectors, summed up
- Higher score = more likely to be a true fact
- Works best for symmetric relations where direction doesn’t matter
How It Works
DistMult (Yang et al., 2015) uses a diagonal bilinear form to score triples. Score: Σ_k (h_k · r_k · t_k). Higher scores indicate more plausible triples. Training uses logistic loss with L2 regularization on embeddings.
DistMult is inherently symmetric: score(h, r, t) = score(t, r, h), making it best suited for symmetric relations like “similar_to” or “married_to”.
Constructors
| Constructor | Summary |
|---|---|
DistMultEmbedding() |
Properties
| Property | Summary |
|---|---|
bool IsDistanceBased { get; } |