User recommendation preference model based on attention enhanced knowledge perception

Take notes
article
AKUPM: Attention-Enhanced Knowledge-Aware User Preference Model for Recommendation

Notes are simple . Google the original link ！ Too lazy to post

motivation ：
The recommended results may be affected by a large number of unrelated entities , To solve this problem , Explore relationships between entities .

The interaction between entities can be divided into ： Interaction between entities （ The importance of entities when representing users ）、 Intra entity interaction （ Different characteristics of entities under different relationships ）.

Strategy ：
1, Take advantage of the self attention network , Capture the interaction between entities by learning the appropriate importance of each entity to the user .
2, By projecting each entity into its connected relational space to obtain appropriate features （embedding）, So as to model the interaction between entities .
By doing so ,AKUPM Be able to find the most relevant part of the merged entity ( That is, filter out irrelevant entities ).

frame ：

Method ：
1,entity propagation
To filter the noise , We propose to merge entities propagated from the user's click history in the relationship of knowledge graph , Make each merged entity relevant to the user .
polymerization k Skip neighbor's entity information

2,Entity Representation
TransR

3,Attention-based User Representation
Self attention mechanism