Heterogeneous graph attention classifier (HGAT) — heterogeneous

Implements Linmei et al.'s (2019) heterogeneous graph convolution and dual-level attention. Node types retain separate input feature spaces and transformation matrices. For every target node, type-level attention first weighs the available neighboring information types; node-level attention then weighs individual neighbors before their type-specific transformed representations are combined. This is HGAT for heterogeneous information networks, distinct from Ding et al.'s document-hypergraph text_hypergat().

Usage

heterogeneous_hgat(
  adjacency,
  node_types,
  features,
  labels,
  hidden = 128L,
  layers = 2L,
  epochs = 200L,
  lr = 0.005,
  weight_decay = 5e-04,
  dropout = 0.5,
  validation = 0.1,
  seed = 1L,
  verbose = FALSE
)

Arguments

adjacency: Symmetric non-negative adjacency matrix with identical unique row and column names.
node_types: Named character vector assigning every node a type.
features: Named list with one numeric feature matrix per type. Rows are named nodes of that type; feature widths may differ across types.
labels: Named character labels for a subset of nodes.
hidden: Common hidden dimension.
layers: Number of dual-attention layers (paper default 2).
epochs, lr, weight_decay, dropout, validation, seed, verbose: Training controls as in hg_neural().

Value

A prediction data.frame for every heterogeneous-network node, with training history and node types attached as attributes.

References

Linmei, H., Yang, T., Shi, C., Ji, H., & Li, X. (2019). Heterogeneous graph attention networks for semi-supervised short text classification. EMNLP-IJCNLP 2019, 4821–4830.