Preference perception recommendation method based on deep reinforcement learning
The invention provides a preference perception recommendation method based on deep reinforcement learning, which learns fine-grained user-entity-relationship preference information in a knowledge graph, and constructs a heterogeneous weight graph of a user for assisting recommendation. Firstly, a de...
Saved in:
Main Authors | , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
04.04.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention provides a preference perception recommendation method based on deep reinforcement learning, which learns fine-grained user-entity-relationship preference information in a knowledge graph, and constructs a heterogeneous weight graph of a user for assisting recommendation. Firstly, a deep reinforcement learning model is used to construct a path network between historical items of a user in a knowledge graph; and then, iteratively diffusing and representing historical project nodes in the knowledge graph into clusters, constructing paths between the clusters, and mining potential relationships between the nodes in different clusters. The reinforcement learning formulates a corresponding feedback reward according to the hierarchical propagation path, and learns the weight of the edge based on the expected return value of each user-entity-relationship, thereby generating a weight knowledge graph with fine-grained user preferences. And finally, aggregating the candidate items and the high-order repre |
---|---|
Bibliography: | Application Number: CN202211415777 |