Context-aware Graph Embedding for Session-based News Recommendation
RecSys 2020
Focus on session-based news recommendation
Abstract
Existing methods ignore the semantic-level structural information among news articles and do not explore external knowledge sources.
- propose a novel Context-aware Graph Embedding(CAGE) framework for session-based news recommendation
- builds an auxiliary KG to enrich the semantic meaning of entities, and further refines the article embeddings by graph convolutional networks
Model
- extract entities from news articles
- build sub-KG from open KG
- obtain article content embeddings through word embedding and CNN
- concatenate
- feature refinement by GCN
- RNN to obtain session-level embeddings
- predictor
Textual-Level Article Embedding
Semantic-Level Embedding
Refining Article Embeddings with GNN
concate article_emb, user_feature_emb and semantic_emb
article emb:
p: user attributes
- articles with similar concepts shall be close in the embedding space
so contrust article-level graphs and employ GNN to refine embs.
- articles are nodes
- pair-wise similarity of article embeddings are weights on edges
- remove edges with small similarity
- two-layer GNN
D: degree matrix
A: adjacency matrix
Framework
Experments
- Dataset
Adressa + wididata KG