Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
80 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

DiffKG: Knowledge Graph Diffusion Model for Recommendation (2312.16890v1)

Published 28 Dec 2023 in cs.IR

Abstract: Knowledge Graphs (KGs) have emerged as invaluable resources for enriching recommendation systems by providing a wealth of factual information and capturing semantic relationships among items. Leveraging KGs can significantly enhance recommendation performance. However, not all relations within a KG are equally relevant or beneficial for the target recommendation task. In fact, certain item-entity connections may introduce noise or lack informative value, thus potentially misleading our understanding of user preferences. To bridge this research gap, we propose a novel knowledge graph diffusion model for recommendation, referred to as DiffKG. Our framework integrates a generative diffusion model with a data augmentation paradigm, enabling robust knowledge graph representation learning. This integration facilitates a better alignment between knowledge-aware item semantics and collaborative relation modeling. Moreover, we introduce a collaborative knowledge graph convolution mechanism that incorporates collaborative signals reflecting user-item interaction patterns, guiding the knowledge graph diffusion process. We conduct extensive experiments on three publicly available datasets, consistently demonstrating the superiority of our DiffKG compared to various competitive baselines. We provide the source code repository of our proposed DiffKG model at the following link: https://github.com/HKUDS/DiffKG.

PDF HTML Abstract

Introducing DiffKG: Enhancing Recommendation Systems through Knowledge Graph Diffusion Models

Knowledge-aware Recommendation Enhanced by Diffusion Models

The field of recommendation systems has witnessed rapid advancements, spearheaded by the integration of knowledge graphs (KGs) to tackle the perpetual challenge of sparse user-item interactions. In this paper, we introduce the DiffKG framework, a novel approach that harnesses the power of generative diffusion models to refine knowledge graph (KG) representations for the specific task of recommendation. This technique addresses the critical issue of noise and irrelevant information in KGs, which can adversely affect recommendation quality.

Key Innovations of DiffKG

The DiffKG framework is distinguished by three pivotal contributions to the domain of knowledge-aware recommendation systems:

  1. Task-Specific Knowledge Graph Optimization: Unlike conventional methods that utilize KGs as-is, the DiffKG model applies a diffusion process to iteratively corrupt and reconstruct the KG. This process effectively filters out irrelevant entity relationships, preserving only those that are pertinent to the recommendation task.
  2. Generative Diffusion Model Integration: At its core, the DiffKG utilizes a generative diffusion paradigm to iteratively model the distribution of relevant KG relationships. This innovative step enables the encoding of user preferences more accurately by ensuring that only task-relevant KG information is considered in the recommendation process.
  3. Collaborative Knowledge Convolution Mechanism: To further align the revised KG with user-item interaction patterns, DiffKG introduces a collaborative knowledge graph convolution (CKGC) mechanism. This component enhances the KG diffusion process with collaborative signals, ensuring the distilled KG is optimally aligned with the underlying recommendation tasks.

Comprehensive Evaluation and Insights

Extensive experiments conducted on public datasets across various domains (music, news, e-commerce) validate the superiority of DiffKG over established baselines, including both traditional collaborative filtering models and contemporary KG-enhanced recommenders. These results underscore the framework's capability to effectively mitigate the impacts of data sparsity and noise, two prevalent challenges in recommendation systems.

Addressing Data Sparsity and Noise

The evaluation particularly highlights DiffKG's adeptness at navigating the field of sparse user interaction data and noisy KG inputs. By generating task-specific KGs that seamlessly integrate with collaborative filtering processes, DiffKG exhibits remarkable resilience against these pervasive issues, offering refined and relevant recommendations even under challenging conditions.

Future Directions

The promising outcomes of this paper pave the way for further exploration into the synergistic potential of diffusion models and KGs in enhancing recommendation systems. Future research could delve into optimizing the diffusion process for dynamic KGs or extending the framework to incorporate temporal dynamics in user-item interactions. Another intriguing avenue involves investigating the interpretability of recommendations provided by DiffKG, offering insights into the decision-making process and the role of KG relations in shaping user preference modeling.

Concluding Remarks

DiffKG stands out as a significant stride forward in the evolution of knowledge-aware recommendation systems. By judiciously integrating generative diffusion models with knowledge graph learning, the framework sets a new benchmark in addressing data quality challenges prevalent in recommendation scenarios. As we move forward, the principles and methodologies underscored by DiffKG will undoubtedly inspire further innovations, pushing the boundaries of personalized recommendation systems.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (45)
  1. Graph convolutional matrix completion. arXiv preprint arXiv:1706.02263 (2017).
  2. Unifying knowledge graph learning and recommendation: Towards a better understanding of user preferences. In WWW. 151–161.
  3. Heterogeneous graph contrastive learning for recommendation. In WSDM. 544–552.
  4. DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models. In ICLR.
  5. Vector quantized diffusion model for text-to-image synthesis. In CVPR. 10696–10706.
  6. Lightgcn: Simplifying and powering graph convolution network for recommendation. In SIGIR. 639–648.
  7. Neural collaborative filtering. In WWW. 173–182.
  8. Denoising diffusion probabilistic models. NeurIPS 33 (2020), 6840–6851.
  9. Cascaded diffusion models for high fidelity image generation. JLMR 23, 1 (2022), 2249–2281.
  10. Graphgdp: Generative diffusion processes for permutation invariant graph generation. In ICDM. IEEE, 201–210.
  11. Advances in collaborative filtering. Recommender systems handbook (2021), 91–142.
  12. Walid Krichene and Steffen Rendle. 2020. On sampled metrics for item recommendation. In KDD. 1748–1757.
  13. Multi-Intention Oriented Contrastive Learning for Sequential Recommendation. In WSDM. 411–419.
  14. Diffusion-lm improves controllable text generation. NeurIPS 35 (2022), 4328–4343.
  15. Variational autoencoders for collaborative filtering. In WWW. 689–698.
  16. Repaint: Inpainting using denoising diffusion probabilistic models. In CVPR. 11461–11471.
  17. Sparsity and noise: Where knowledge graph embeddings fall short. In EMNLP. 1751–1756.
  18. Representation Learning with Large Language Models for Recommendation. arXiv preprint arXiv:2310.15950 (2023).
  19. Disentangled Contrastive Collaborative Filtering. arXiv preprint arXiv:2305.02759 (2023).
  20. Distillation-Enhanced Graph Masked Autoencoders for Bundle Recommendation. In SIGIR. 1660–1669.
  21. BPR: Bayesian personalized ranking from implicit feedback. arXiv preprint arXiv:1205.2618 (2012).
  22. Deep unsupervised learning using nonequilibrium thermodynamics. In ICML. PMLR, 2256–2265.
  23. BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer. In CIKM. 1441–1450.
  24. Joint knowledge pruning and recurrent graph convolution for news recommendation. In SIGIR. 51–60.
  25. Graph attention networks. arXiv preprint arXiv:1710.10903 (2017).
  26. DiGress: Discrete Denoising diffusion for graph generation. In ICLR.
  27. Label-free distant supervision for relation extraction via knowledge graph embedding. In EMNLP. 2246–2255.
  28. DKN: Deep knowledge-aware network for news recommendation. In WWW. 1835–1844.
  29. Knowledge-aware graph neural networks with label smoothness regularization for recommender systems. In KDD. 968–977.
  30. Knowledge graph convolutional networks for recommender systems. In WWW. 3307–3313.
  31. Diffusion Recommender Model. In SIGIR.
  32. Kgat: Knowledge graph attention network for recommendation. In KDD. 950–958.
  33. Learning intents behind interactions with knowledge graph for recommendation. In WWW. 878–887.
  34. Explainable reasoning over knowledge graphs for recommendation. In AAAI, Vol. 33. 5329–5336.
  35. Multi-Modal Self-Supervised Learning for Recommendation. In WWW. 790–800.
  36. LLMRec: Large Language Models with Graph Augmentation for Recommendation. arXiv preprint arXiv:2311.00423 (2023).
  37. Self-supervised graph learning for recommendation. In SIGIR. 726–735.
  38. Debiased Contrastive Learning for Sequential Recommendation. In WWW. 1063–1073.
  39. Knowledge graph contrastive learning for recommendation. In SIGIR. 1434–1443.
  40. Self-supervised learning for large-scale item recommendations. In CIKM. 4321–4330.
  41. Personalized entity recommendation: A heterogeneous information network approach. In WSDM. 283–292.
  42. Collaborative knowledge base embedding for recommender systems. In KDD. 353–362.
  43. Kb4rec: A data set for linking knowledge bases with recommender systems. Data Intelligence 1, 2 (2019), 121–136.
  44. Multi-level cross-view contrastive learning for knowledge-aware recommender system. In SIGIR. 1358–1368.
  45. Improving knowledge-aware recommendation with multi-level interactive contrastive learning. In CIKM. 2817–2826.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Yangqin Jiang (6 papers)
  2. Yuhao Yang (23 papers)
  3. Lianghao Xia (65 papers)
  4. Chao Huang (244 papers)
Citations (27)
View on Semantic Scholar
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. GitHub - HKUDS/DiffKG: [WSDM'2024 Oral] "DiffKG: Knowledge Graph Diffusion Model for Recommendation" (113 stars)
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets