System and method for explainable embedding-based recommendation system
A method includes obtaining, by an electronic device, an interpretation hierarchy generated based on a knowledge graph and behavioral data. The method also includes performing, by the electronic device, graph convolution operations on the interpretation hierarchy to generate one or more embeddings. The method further includes generating, by the electronic device, a recommendation based at least in part on associations between the one or more embeddings. In addition, the method includes providing, by the electronic device, an explanation corresponding to the recommendation.
1 . An electronic device comprising:
at least one memory configured to store a database; and
at least one processor coupled to the at least one memory, the at least one processor configured to:
obtain an interpretation hierarchy generated based on a knowledge graph and behavioral data;
train a machine learning model, wherein the at least one processor is configured to:
collect low-level node embeddings based on the interpretation hierarchy;
assign the low-level node embeddings to different factors including node type, semantic factors of items, and semantic factors of knowledge graph entities;
combine the low-level node embeddings for each factor;
generate, using the combined low-level node embeddings, a plurality of embeddings including user embeddings, item embeddings, and knowledge graph entity embeddings, wherein the user embeddings comprise a concatenation of a user-based embedding, an item-based embedding, and an entity-based embedding; and
update the machine learning model based on the plurality of embeddings; and
provide, using the machine learning model, an output and a representation of the output.
2 . The electronic device of claim 1 , wherein the user embeddings, the item embeddings, and the knowledge graph entity embeddings are generated by graph convolution operations and disposed on different layers within the interpretation hierarchy.
3 . The electronic device of claim 2 , wherein the item embeddings comprise a concatenation of another item-based embedding and another entity-based embedding.
4 . The electronic device of claim 1 , wherein the interpretation hierarchy connects to multi-level knowledge graph entities.
5 . The electronic device of claim 1 , wherein:
the at least one processor is further configured to provide a prediction; and
the prediction is a function of an item based embedding component and an entity based embedding component.
6 . The electronic device of claim 1 , wherein, to generate the plurality of embeddings, the at least one processor is configured to:
perform a first convolution operation to generate the item entity embeddings and the user embeddings via separating information from different types of nodes; and
perform a second convolution operation to aggregate the item embeddings and the user embeddings by introducing the semantic factors of the items and the semantic factors of the knowledge graph entities.
7 . The electronic device of claim 1 , wherein the representation of the output is one or more of a graphic representation, an image representation, or plain text.
8 . A method comprising:
obtaining, by an electronic device, an interpretation hierarchy generated based on a knowledge graph and behavioral data;
training a machine learning model by:
collecting low-level node embeddings based on the interpretation hierarchy;
assigning the low-level node embeddings to different factors including node type, semantic factors of items, and semantic factors of knowledge graph entities;
combining the low-level node embeddings for each factor;
generating, using the combined low-level node embeddings, a plurality of embeddings including user embeddings, item embeddings, and knowledge graph entity embeddings, wherein the user embeddings comprise a concatenation of a user-based embedding, an item-based embedding, and an entity-based embedding; and
updating the machine learning model based on the plurality of embeddings; and
providing, using the machine learning model, an output and a representation of the output.
9 . The method of claim 8 , wherein the user embeddings, the item embeddings, and the knowledge graph entity embeddings are generated by graph convolution operations and disposed on different layers within the interpretation hierarchy.
10 . The method of claim 9 , wherein the item embeddings comprise a concatenation of another item-based embedding and another entity-based embedding.
11 . The method of claim 8 , wherein the interpretation hierarchy connects to multi-level knowledge graph entities.
12 . The method of claim 8 , further comprising:
providing, by the electronic device, a prediction;
wherein the prediction is a function of an item based embedding component and an entity based embedding component.
13 . The method of claim 8 , wherein generating the plurality of embeddings comprises:
performing a first convolution operation to generate the item entity embeddings and the user embeddings via separating information from different types of nodes; and
performing a second convolution operation to aggregate the item embeddings and the user embeddings by introducing the semantic factors of the items and the semantic factors of the knowledge graph entities.
14 . The method of claim 8 , wherein the representation of the output is one or more of a graphic representation, an image representation, or plain text.
15 . A non-transitory machine-readable medium comprising instructions that, when executed by at least one processor of an electronic device, cause the at least one processor to:
obtain an interpretation hierarchy generated based on a knowledge graph and behavioral data;
train a machine learning model, wherein the instructions when executed cause the at least one processor to:
collect low-level node embeddings based on the interpretation hierarchy;
assign the low-level node embeddings to different factors including node type, semantic factors of items, and semantic factors of knowledge graph entities;
combine the low-level node embeddings for each factor;
generate, using the combined low-level node embeddings, a plurality of embeddings including user embeddings, item embeddings, and knowledge graph entity embeddings, wherein the user embeddings comprise a concatenation of a user-based embedding, an item-based embedding, and an entity-based embedding; and
update the machine learning model based on the plurality of embeddings; and
provide, using the machine learning model, an output and a representation of the output.
16 . The non-transitory machine-readable medium of claim 15 , wherein:
the user embeddings, the item embeddings, and the knowledge graph entity embeddings are generated by graph convolution operations and disposed on different layers within the interpretation hierarchy; and
the item embeddings comprise a concatenation of another item-based embedding and another entity-based embedding.
17 . The non-transitory machine-readable medium of claim 15 , wherein the interpretation hierarchy connects to multi-level knowledge graph entities.
18 . The non-transitory machine-readable medium of claim 17 , wherein:
the instructions when executed further cause the at least one processor to provide a prediction; and
the prediction is a function of an item based embedding component and an entity based embedding component.
19 . The non-transitory machine-readable medium of claim 15 , wherein the instructions that when executed cause the at least one processor to generate the plurality of embeddings comprise instructions that when executed cause the at least one processor to:
perform a first convolution operation to generate the item entity embeddings and the user embeddings via separating information from different types of nodes; and
perform a second convolution operation to aggregate the item embeddings and the user embeddings by introducing the semantic factors of the items and the semantic factors of the knowledge graph entities.
20 . The non-transitory machine-readable medium of claim 15 , wherein the representation of the output is one or more of a graphic representation, an image representation, or plain text.