End-to-End Graph-Sequential Representation Learning for Accurate Recommendations

Abstract

The proposed MRGSRec model consists of nine blocks, colored in yellow and red: (i) Item Embedding Layer and (ii) User Embedding Layer that encodes input item and user IDs. (iii) Sequential Encoder and (iv) Graph Encoder that aggregate information leveraging different data representations. (v) Fusing Layer capable of mixing representations from both encoders. (vi) Local Objective, (vii) Global Objective, (viii) Fusing Objective, and (ix) Contrastive Objective used for model training