We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Information Retrieval

Title: Pre-training Graph Neural Network for Cross Domain Recommendation

Abstract: A recommender system predicts users' potential interests in items, where the core is to learn user/item embeddings. Nevertheless, it suffers from the data-sparsity issue, which the cross-domain recommendation can alleviate. However, most prior works either jointly learn the source domain and target domain models, or require side-features. However, jointly training and side features would affect the prediction on the target domain as the learned embedding is dominated by the source domain containing bias information. Inspired by the contemporary arts in pre-training from graph representation learning, we propose a pre-training and fine-tuning diagram for cross-domain recommendation. We devise a novel Pre-training Graph Neural Network for Cross-Domain Recommendation (PCRec), which adopts the contrastive self-supervised pre-training of a graph encoder. Then, we transfer the pre-trained graph encoder to initialize the node embeddings on the target domain, which benefits the fine-tuning of the single domain recommender system on the target domain. The experimental results demonstrate the superiority of PCRec. Detailed analyses verify the superiority of PCRec in transferring information while avoiding biases from source domains.
Subjects: Information Retrieval (cs.IR); Machine Learning (cs.LG)
Cite as: arXiv:2111.08268 [cs.IR]
  (or arXiv:2111.08268v1 [cs.IR] for this version)

Submission history

From: Chen Wang [view email]
[v1] Tue, 16 Nov 2021 07:34:42 GMT (162kb,D)

Link back to: arXiv, form interface, contact.