recbole.data.kg_dataset¶

class recbole.data.dataset.kg_dataset.KnowledgeBasedDataset(config)[source]¶

Bases: Dataset

KnowledgeBasedDataset is based on Dataset, and load .kg and .link additionally.

Entities are remapped together with item_id specially. All entities are remapped into three consecutive ID sections.

virtual entities that only exist in interaction data.
entities that exist both in interaction data and kg triplets.
entities only exist in kg triplets.

It also provides several interfaces to transfer .kg features into coo sparse matrix, csr sparse matrix, DGL.Graph or PyG.Data.

head_entity_field¶

The same as config['HEAD_ENTITY_ID_FIELD'].

Type:: str

tail_entity_field¶

The same as config['TAIL_ENTITY_ID_FIELD'].

Type:: str

relation_field¶

The same as config['RELATION_ID_FIELD'].

Type:: str

entity_field¶

The same as config['ENTITY_ID_FIELD'].

Type:: str

kg_feat¶

Internal data structure stores the kg triplets. It’s loaded from file .kg.

Type:: pandas.DataFrame

item2entity¶

Dict maps item_id to entity, which is loaded from file .link.

Type:: dict

entity2item¶

Dict maps entity to item_id, which is loaded from file .link.

Type:: dict

Note

entity_field doesn’t exist exactly. It’s only a symbol, representing entity features.

[UI-Relation] is a special relation token.

ckg_graph(form='coo', value_field=None)[source]¶

Get graph or sparse matrix that describe relations of CKG, which combines interactions and kg triplets into the same graph.

Item ids and entity ids are added by user_num temporally.

For an edge of <src, tgt>, graph[src, tgt] = 1 if value_field is None, else graph[src, tgt] = self.kg_feat[self.relation_field][src, tgt] or graph[src, tgt] = [UI-Relation].

Currently, we support graph in DGL and PyG, and two type of sparse matrices, coo and csr.

Parameters:

form (str, optional) – Format of sparse matrix, or library of graph data structure. Defaults to coo.
value_field (str, optional) – self.relation_field or None, Defaults to None.

Returns:

Graph / Sparse matrix of kg triplets.

property entities¶: Returns: numpy.ndarray: List of entity id, including virtual entities.

property entity_num¶

Get the number of different tokens of entities, including virtual entities.

Returns:: Number of different tokens of entities, including virtual entities.
Return type:: int

property head_entities¶: Returns: numpy.ndarray: List of head entities of kg triplets.

kg_graph(form='coo', value_field=None)[source]¶

Get graph or sparse matrix that describe relations between entities.

For an edge of <src, tgt>, graph[src, tgt] = 1 if value_field is None, else graph[src, tgt] = self.kg_feat[value_field][src, tgt].

Currently, we support graph in DGL and PyG, and two type of sparse matrices, coo and csr.

Parameters:

form (str, optional) – Format of sparse matrix, or library of graph data structure. Defaults to coo.
value_field (str, optional) – edge attributes of graph, or data of sparse matrix, Defaults to None.

Returns:

Graph / Sparse matrix of kg triplets.

property relation_num¶

Get the number of different tokens of self.relation_field.

Returns:: Number of different tokens of self.relation_field.
Return type:: int

property relations¶: Returns: numpy.ndarray: List of relations of kg triplets.

property tail_entities¶: Returns: numpy.ndarray: List of tail entities of kg triplets.