Back

Graph Neural Networks for Social Recommendation

Wenqi FanYao MaQing LiYuan HeEric ZhaoJiliang TangDawei Yin

article2019WWW2,413 citations

Develops GraphRec, a graph neural network framework that captures user-item interactions, rating opinions, and social ties with varying relationship strengths to improve social recommendation accuracy.

Listen

Online platforms increasingly rely on recommender systems to guide user choices, and incorporating social connections has become a primary way to refine these recommendations. However, existing social recommendation methods struggle to effectively combine user-to-user social networks with user-to-item interaction networks. They also frequently overlook the specific opinions (such as rating scores) expressed during interactions and treat all social relationships as having equal influence, despite real-world social ties varying widely in strength.

This article introduces GraphRec, a graph neural network framework designed to improve rating prediction accuracy by coherently modeling social graphs and user-item interaction graphs. The article sets out to demonstrate how graph neural networks can jointly integrate user opinions and distinguish between varying strengths of social relationships.

To evaluate this framework, the authors conducted experimental evaluations on two real-world benchmark datasets from online consumer platforms: Ciao (comprising over 7,300 users and 283,000 ratings) and Epinions (comprising over 18,000 users and 764,000 ratings). GraphRec was tested across varying training data splits against eight baseline models, spanning standard matrix factorization, conventional social recommendation models, and advanced deep learning systems. Performance was assessed using standard predictive error metrics, specifically Mean Absolute Error and Root Mean Square Error.

The findings establish that GraphRec consistently outperforms all baseline models across both datasets. For instance, in 80% training data splits, GraphRec achieved the lowest error rates on Ciao (RMSE of 0.9794, MAE of 0.7387) and Epinions (RMSE of 1.0631, MAE of 0.8168). Component analysis revealed that removing explicit opinion embeddings caused performance to drop significantly, increasing prediction error by roughly 2.6% to 5.8% across metrics. Furthermore, eliminating attention mechanisms—which assign individualized weights to strong versus weak social ties and varying item interactions—noticeably degraded accuracy. Finally, testing embedding sizes showed that performance peaked around a moderate dimension (size 64), whereas overly large embeddings increased computational complexity and reduced accuracy.

These results demonstrate that accurately capturing social tie strength and explicit user feedback is critical for recommendation performance. In practical terms, even marginal reductions in prediction error translate to substantially higher quality in top recommendations. This improved accuracy can drive user engagement, enhance conversion rates, and optimize algorithmic decision-making across e-commerce and social platforms without requiring cumbersome external data sources.

Organizations operating social or e-commerce platforms should consider adopting graph neural network architectures that utilize attention mechanisms to capture diverse interaction weights and relationship strengths. Technical teams should carefully tune embedding dimensions to balance recommendation accuracy against computational and infrastructure overhead. Before full deployment, stakeholders should conduct pilot testing to determine the ideal operational parameters for their specific workload.

The main limitations of the article are its evaluation on static datasets and its exclusive focus on user ratings and social links, omitting dynamic temporal changes and richer metadata such as user demographics or item text descriptions. Nonetheless, given the rigorous benchmarking across standard datasets, confidence in the reported performance advantages remains high for comparable static recommendation environments.

City University of Hong KongHong Kong Polytechnic UniversityJD.comMichigan State Universitygraph neural networkslatent factorrecommendationgraph neural networkssocial recommendationneural collaborative filtering
arXiv: 1902.07243wenqifan03/GraphRec-WWW19

No sufficiently relevant recommendations were found.

Cover for Graph Neural Networks for Social Recommendation

Abstract

In recent years, Graph Neural Networks (GNNs), which can naturally integrate node information and topological structure, have been demonstrated to be powerful in learning on graph data. These advantages of GNNs provide great potential to advance social recommendation since data in social recommender systems can be represented as user-user social graph and user-item graph; and learning latent factors of users and items is the key. However, building social recommender systems based on GNNs faces challenges. For example, the user-item graph encodes both interactions and their associated opinions; social relations have heterogeneous strengths; users involve in two graphs (e.g., the user-user social graph and the user-item graph). To address the three aforementioned challenges simultaneously, in this paper, we present a novel graph neural network framework (GraphRec) for social recommendations. In particular, we provide a principled approach to jointly capture interactions and opinions in the user-item graph and propose the framework GraphRec, which coherently models two graphs and heterogeneous strengths. Extensive experiments on two real-world datasets demonstrate the effectiveness of the proposed framework GraphRec. Our code is available at \url{this https URL}

Topics

graph neural networkslatent factorrecommendationsocial relationstopological structuresocial graphscollaborative filteringuser informationbipartite graphsdirected graphs

Keywords

graph neural networkssocial recommendationneural collaborative filteringGraphRec modelprobabilistic matrix factorizationuser-item graphsocial graphrating predictionsocial regularizationattention network

Table of Contents

  • 1 Introduction
  • 2 The Proposed Framework
  • 2.1 Definitions and Notations
  • 2.2 An Overview of the Proposed Framework
  • 2.3 User Modeling
  • 2.4 Item Modeling
  • 2.5 Rating Prediction
  • 2.6 Model Training
  • 3 Experiment
  • 3.1 Experimental Settings
  • 3.1.1 Datasets
  • 3.1.2 Evaluation Metrics
  • 3.1.3 Baselines
  • 3.1.4 Parameter Settings
  • 3.2 Performance Comparison of Recommender Systems
  • 3.3 Model Analysis
  • 3.3.1 Effect of Social Network and User Opinions
  • 3.3.2 Effect of Attention Mechanisms
  • 3.3.3 Effect of Embedding Size
  • 4 Related work
  • 5 Conclusion and Future work
  • References

Knowls

  1. Knowl 1 — Item-Space User Representation Learning with Opinion-Aware Attention

    model/method

    In GraphRec, a user's preference over items is captured through an item-space user latent factor hiI∈Rdh_i^I \in \mathbb{R}^d for user ui∈Uu_i \in U, learned from the set of items C(i)C(i) that uiu_i has interacted with in the user-item bipartite graph.

    To explicitly account for the rating score (opinion) r∈{1,2,…,5}r \in \{1, 2, \dots, 5\} that user uiu_i assigned to an item va∈C(i)v_a \in C(i), an opinion embedding er∈Rde_r \in \mathbb{R}^d is concatenated with the item embedding qa∈Rdq_a \in \mathbb{R}^d and passed through a multi-layer perceptron gvg_v to form an opinion-aware interaction representation:

    xia=gv([qa⊕er])x_{ia} = g_v([q_a \oplus e_r])

    where ⊕\oplus denotes vector concatenation.

    Because different interactions contribute unequally to a user's latent preference, GraphRec applies a two-layer neural attention network that conditions on xiax_{ia} and the target user embedding pi∈Rdp_i \in \mathbb{R}^d:

    αia∗=w2Tσ(W1[xia⊕pi]+b1)+b2\alpha^*_{ia} = w_2^T \sigma(W_1 [x_{ia} \oplus p_i] + b_1) + b_2

    αia=exp⁡(αia∗)∑a′∈C(i)exp⁡(αia′∗)\alpha_{ia} = \frac{\exp(\alpha^*_{ia})}{\sum_{a' \in C(i)} \exp(\alpha^*_{ia'})}

    where W1W_1, b1b_1, w2w_2, and b2b_2 are learnable projection parameters, and σ\sigma is the ReLU activation function.

    The item-space user latent factor hiIh_i^I is then aggregated as:

    hiI=σ(W⋅∑a∈C(i)αiaxia+b)h_i^I = \sigma\left(W \cdot \sum_{a \in C(i)} \alpha_{ia} x_{ia} + b\right)

    where WW and bb are the aggregation weight matrix and bias vector.

  2. Knowl 2 — Social-Space User Representation Learning with Tie-Strength Attention

    model/method

    In GraphRec, social relations from the user-user social graph T∈Rn×nT \in \mathbb{R}^{n \times n} are integrated to compute a social-space user latent factor hiS∈Rdh_i^S \in \mathbb{R}^d for user uiu_i. To model heterogeneous tie strengths among social connections, the model aggregates the item-space user latent factors hoI∈Rdh_o^I \in \mathbb{R}^d of direct neighbors o∈N(i)o \in N(i) using an attention mechanism.

    The social attention weight βio\beta_{io}, representing the influence strength of neighbor uou_o on user uiu_i, is computed by conditioning on the neighbor's item-space latent factor hoIh_o^I and the target user's embedding pi∈Rdp_i \in \mathbb{R}^d:

    βio∗=w2Tσ(W1[hoI⊕pi]+b1)+b2\beta^*_{io} = w_2^T \sigma(W_1 [h_o^I \oplus p_i] + b_1) + b_2

    βio=exp⁡(βio∗)∑o′∈N(i)exp⁡(βio′∗)\beta_{io} = \frac{\exp(\beta^*_{io})}{\sum_{o' \in N(i)} \exp(\beta^*_{io'})}

    where ⊕\oplus denotes vector concatenation, σ\sigma is the ReLU activation, and W1,b1,w2,b2W_1, b_1, w_2, b_2 are learnable parameters.

    The social-space user latent factor is obtained via:

    hiS=σ(W⋅∑o∈N(i)βiohoI+b)h_i^S = \sigma\left(W \cdot \sum_{o \in N(i)} \beta_{io} h_o^I + b\right)

    To construct the comprehensive user latent factor hi∈Rdh_i \in \mathbb{R}^d, the item-space latent factor hiIh_i^I and social-space latent factor hiSh_i^S are concatenated and transformed through a multi-layer perceptron (MLP):

    c1=[hiI⊕hiS]c_1 = [h_i^I \oplus h_i^S]

    cl=σ(Wlcl−1+bl),l=2,…,Lc_l = \sigma(W_l c_{l-1} + b_l), \quad l = 2, \dots, L

    hi=cLh_i = c_L

  3. Knowl 3 — Item Representation Learning via Opinion-Aware User Aggregation

    model/method

    In GraphRec, the latent factor zj∈Rdz_j \in \mathbb{R}^d for an item vj∈Vv_j \in V is learned by aggregating information from the set of users B(j)B(j) who interacted with vjv_j, taking into account the varying opinions expressed in each interaction.

    For an interaction between user ut∈B(j)u_t \in B(j) and item vjv_j with rating score rr, an opinion-aware interaction user representation fjtf_{jt} is constructed using a multi-layer perceptron gug_u that fuses the user embedding pt∈Rdp_t \in \mathbb{R}^d and the opinion embedding er∈Rde_r \in \mathbb{R}^d:

    fjt=gu([pt⊕er])f_{jt} = g_u([p_t \oplus e_r])

    To account for the heterogeneous influence of different users on item representation, a two-layer neural attention network scores the contribution of user utu_t relative to item embedding qj∈Rdq_j \in \mathbb{R}^d:

    μjt∗=w2Tσ(W1[fjt⊕qj]+b1)+b2\mu^*_{jt} = w_2^T \sigma(W_1 [f_{jt} \oplus q_j] + b_1) + b_2

    μjt=exp⁡(μjt∗)∑t′∈B(j)exp⁡(μjt′∗)\mu_{jt} = \frac{\exp(\mu^*_{jt})}{\sum_{t' \in B(j)} \exp(\mu^*_{jt'})}

    The final item latent factor zjz_j is obtained by attentive aggregation:

    zj=σ(W⋅∑t∈B(j)μjtfjt+b)z_j = \sigma\left(W \cdot \sum_{t \in B(j)} \mu_{jt} f_{jt} + b\right)

    where σ\sigma is ReLU, and WW and bb are trainable transformation parameters.

  4. Knowl 4 — Rating Prediction and Optimization Objective in GraphRec

    model/method

    GraphRec performs rating prediction by concatenating the final user latent factor hi∈Rdh_i \in \mathbb{R}^d and item latent factor zj∈Rdz_j \in \mathbb{R}^d into a joint representation vector, which is fed into a multi-layer perceptron (MLP):

    g1=[hi⊕zj]g_1 = [h_i \oplus z_j]

    gl=σ(Wlgl−1+bl),l=2,…,L−1g_l = \sigma(W_l g_{l-1} + b_l), \quad l = 2, \dots, L-1

    rij′=wTgL−1r'_{ij} = w^T g_{L-1}

    where rij′r'_{ij} is the predicted rating from user uiu_i to item vjv_j, σ\sigma is ReLU, and Wl,bl,wW_l, b_l, w are trainable parameters.

    The model parameters are optimized end-to-end on the set of observed user-item ratings O={(ui,vj)∣rij≠0}\mathcal{O} = \{(u_i, v_j) \mid r_{ij} \neq 0\} using mean squared error loss:

    Loss=12∣O∣∑(ui,vj)∈O(rij′−rij)2\text{Loss} = \frac{1}{2 |\mathcal{O}|} \sum_{(u_i, v_j) \in \mathcal{O}} (r'_{ij} - r_{ij})^2

    where rijr_{ij} is the ground-truth rating. The network embeddings (pi,qj,erp_i, q_j, e_r) and layer weights are randomly initialized from a Gaussian distribution N(0,0.1)\mathcal{N}(0, 0.1) and optimized using RMSprop. Dropout is applied during training to prevent overfitting.

  5. Knowl 5 — Empirical Rating Prediction Performance Comparison

    data/table

    GraphRec was evaluated against matrix factorization baselines (PMF, SoRec, SoReg, SocialMF, TrustMF), neural collaborative filtering (NeuMF), and neural/graph-based social recommendation models (DeepSoR, GCMC+SN) on the Ciao and Epinions benchmark datasets under 60% and 80% training set splits using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

    Dataset Split Metric NeuMF GCMC+SN GraphRec
    Ciao 60% MAE 0.8251 0.7697 0.7540
    Ciao 60% RMSE 1.0824 1.0221 1.0093
    Ciao 80% MAE 0.8062 0.7526 0.7387
    Ciao 80% RMSE 1.0617 0.9931 0.9794
    Epinions 60% MAE 0.9097 0.8602 0.8441
    Epinions 60% RMSE 1.1645 1.1004 1.0878
    Epinions 80% MAE 0.9072 0.8590 0.8168
    Epinions 80% RMSE 1.1476 1.0711 1.0631

    Full baseline results (PMF, SoRec, SoReg, SocialMF, TrustMF, DeepSoR):

    • Ciao (60%): PMF (MAE: 0.9520, RMSE: 1.1967), SoRec (MAE: 0.8489, RMSE: 1.0738), SoReg (MAE: 0.8987, RMSE: 1.0947), SocialMF (MAE: 0.8353, RMSE: 1.0592), TrustMF (MAE: 0.7681, RMSE: 1.0543), DeepSoR (MAE: 0.7813, RMSE: 1.0437).
    • Ciao (80%): PMF (MAE: 0.9021, RMSE: 1.1238), SoRec (MAE: 0.8410, RMSE: 1.0652), SoReg (MAE: 0.8611, RMSE: 1.0848), SocialMF (MAE: 0.8270, RMSE: 1.0501), TrustMF (MAE: 0.7690, RMSE: 1.0479), DeepSoR (MAE: 0.7739, RMSE: 1.0316).
    • Epinions (60%): PMF (MAE: 1.0211, RMSE: 1.2739), SoRec (MAE: 0.9086, RMSE: 1.1563), SoReg (MAE: 0.9412, RMSE: 1.1936), SocialMF (MAE: 0.8965, RMSE: 1.1410), TrustMF (MAE: 0.8550, RMSE: 1.1505), DeepSoR (MAE: 0.8520, RMSE: 1.1135).
    • Epinions (80%): PMF (MAE: 0.9952, RMSE: 1.2128), SoRec (MAE: 0.8961, RMSE: 1.1437), SoReg (MAE: 0.9119, RMSE: 1.1703), SocialMF (MAE: 0.8837, RMSE: 1.1328), TrustMF (MAE: 0.8410, RMSE: 1.1395), DeepSoR (MAE: 0.8383, RMSE: 1.0972).

    GraphRec consistently achieved the lowest MAE and RMSE across all splits and datasets, outperforming the strongest baseline GCMC+SN.

  6. Knowl 6 — Ablation Analysis of Social Relations and Opinion Embeddings in GraphRec

    empirical result

    Ablation experiments on GraphRec evaluate the relative contribution of social network information and explicit opinion embeddings:

    1. GraphRec-SN (removing the social network aggregation hiSh_i^S, setting hi=hiIh_i = h_i^I): Performance degraded significantly on both Ciao and Epinions datasets, verifying that user social relations provide complementary information to interaction history for learning user representations.

    2. GraphRec-Opinion (removing opinion embeddings ere_r from interaction vectors xiax_{ia} and fjtf_{jt}): Omitting rating opinion representations led to a substantial decline in prediction accuracy. On average, removing opinions resulted in a relative error increase of 3.50% on Ciao and 2.64% on Epinions in RMSE, and 5.84% on Ciao and 5.02% on Epinions in MAE.

    These results demonstrate that capturing both interaction topology and explicit rating opinions is essential for accurate representation learning in social recommendation.

  7. Knowl 7 — Ablation Analysis of Attention Mechanisms in GraphRec

    empirical result

    To analyze the contribution of the three attention modules in GraphRec, four attention-ablated variants were compared against the full model on Ciao and Epinions:

    • GraphRec-α\alpha: Replaces item attention αia\alpha_{ia} with uniform mean aggregation 1∣C(i)∣\frac{1}{|C(i)|}. Performance drops, indicating that different interacted items have unequal relevance to a user's latent preference.
    • GraphRec-β\beta: Replaces social attention βio\beta_{io} with uniform mean aggregation 1∣N(i)∣\frac{1}{|N(i)|}. Performance decreases significantly, proving that distinguishing strong versus weak social ties is critical for social aggregation.
    • GraphRec-α&β\alpha\&\beta: Replaces both item attention and social attention with mean-pooling, showing further performance degradation in user representation learning.
    • GraphRec-μ\mu: Replaces user attention μjt\mu_{jt} with uniform mean aggregation 1∣B(j)∣\frac{1}{|B(j)|}. Performance degrades, verifying that different users contribute unequally when modeling item latent factors.

    Across all cases, GraphRec with all three attention mechanisms achieved the best MAE and RMSE, confirming that individualized, attention-weighted aggregation is superior to uniform neighbor averaging.

  8. Knowl 8 — Impact of Latent Embedding Dimension on GraphRec Performance

    empirical result

    Evaluating GraphRec across embedding dimensions d∈{8,16,32,64,128,256}d \in \{8, 16, 32, 64, 128, 256\} on Ciao and Epinions demonstrates a non-monotonic performance trend:

    • Increasing the embedding size dd from 8 to 64 steadily decreases both RMSE and MAE, improving representation capacity.
    • Peak performance on both datasets is achieved around d=64d = 64.
    • Increasing dd further to 256 leads to performance degradation (higher RMSE and MAE) due to overfitting and excessive model complexity.

    Selecting an intermediate embedding dimension (d≈64d \approx 64) provides the optimal balance between expressive representational power and generalization capacity.

  9. Knowl 9 — Limitations of GraphRec in Dynamic Modeling and Attribute Integration

    limitation

    GraphRec has two primary stated limitations:

    1. Static Graph Assumption: The framework assumes fixed user-item ratings and static user-user social relations, ignoring the dynamic temporal evolution inherent in real-world user preferences and social interactions.
    2. Absence of Side Attributes: GraphRec relies solely on interaction graphs, rating scores, and social topologies, without integrating rich auxiliary attributes of users (e.g., demographic data) or items (e.g., categories, descriptions, or multimedia content).

Coverage note — No substantial contributed material was omitted. All architectural components (item-space user modeling, social-space user modeling, item modeling, rating prediction), training procedures, benchmark tables, ablation studies (opinions, social relations, attention modules), hyperparameter sensitivity analyses, and stated limitations are fully covered.

References

  1. 1.Rianne van den Berg, Thomas N Kipf, and Max Welling. 2017. Graph convolutional matrix completion. arXiv preprint arXiv:1706.02263 (2017).
  2. 2.Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. 2017. Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine 34, 4 (2017), 18–42.
  3. 3.Chong Chen, Min Zhang, Yiqun Liu, and Shaoping Ma. 2018. Neural Attentional Rating Regression with Review-level Explanations. In Proceedings of the 27th International Conference on World Wide Web. 1583–1592.
  4. 4.Hongshen Chen, Xiaorui Liu, Dawei Yin, and Jiliang Tang. 2017. A survey on dialogue systems: Recent advances and new frontiers. ACM SIGKDD Explorations Newsletter 19, 2 (2017), 25–35.
  5. 5.Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. 2016. Convolutional neural networks on graphs with fast localized spectral filtering. In Advances in Neural Information Processing Systems. 3844–3852.
  6. 6.Shuiguang Deng, Longtao Huang, Guandong Xu, Xindong Wu, and Zhaohui Wu. 2017. On deep learning for trust-aware recommendations in social networks. IEEE transactions on neural networks and learning systems 28, 5 (2017), 1164–1177.
  7. 7.Tyler Derr, Yao Ma, and Jiliang Tang. 2018. Signed Graph Convolutional Networks. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 929–934.
  8. 8.Wenqi Fan, Qing Li, and Min Cheng. 2018. Deep Modeling of Social Relations for Recommendation. In AAAI.
  9. 9.Aditya Grover and Jure Leskovec. 2016. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 855–864.
  10. 10.Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In Advances in Neural Information Processing Systems. 1024–1034.
  11. 11.Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural Collaborative Filtering. In Proceedings of the 26th International Conference on World Wide Web, WWW. 173–182.
  12. 12.Geoffrey Hinton, Li Deng, Dong Yu, George E Dahl, Abdel-rahman Mohamed, Navdeep Jaitly, Andrew Senior, Vincent Vanhoucke, Patrick Nguyen, Tara N Sainath, et al. 2012. Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Processing Magazine 29, 6 (2012), 82–97.
  13. 13.Mohsen Jamali and Martin Ester. 2010. A matrix factorization technique with trust propagation for recommendation in social networks. In Proceedings of the fourth ACM conference on Recommender systems. ACM, 135–142.
  14. 14.Hamid Karimi, Jiliang Tang, and Yanen Li. 2018. Toward End-to-End Deception Detection in Videos. In 2018 IEEE Big Data. IEEE, 1278–1283.
  15. 15.Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with Graph Convolutional Networks. In International Conference on Learning Representations (ICLR).
  16. 16.Yehuda Koren. 2008. Factorization meets the neighborhood: a multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 426–434.
  17. 17.Hao Ma, Haixuan Yang, Michael R Lyu, and Irwin King. 2008. Sorec: social recommendation using probabilistic matrix factorization. In Proceedings of the 17th ACM conference on Information and Knowledge Management. ACM, 931–940.
  18. 18.Hao Ma, Dengyong Zhou, Chao Liu, Michael R Lyu, and Irwin King. 2011. Recommender systems with social regularization. In Proceedings of the fourth ACM international conference on Web Search and Data Mining. ACM, 287–296.
  19. 19.Yao Ma, Suhang Wang, Charu C. Aggarwal, Dawei Yin, and Jiliang Tang. 2019. Multi-dimensional Graph Convolutional Networks. In Proceedings of the 2019 SIAM International Conference on Data Mining(SDM).
  20. 20.Peter V Marsden and Noah E Friedkin. 1993. Network studies of social influence. Sociological Methods & Research 22, 1 (1993), 127–151.
  21. 21.Miller McPherson, Lynn Smith-Lovin, and James M Cook. 2001. Birds of a feather: Homophily in social networks. Annual review of sociology 27, 1 (2001), 415–444.
  22. 22.Federico Monti, Michael Bronstein, and Xavier Bresson. 2017. Geometric matrix completion with recurrent multi-graph neural networks. In Advances in Neural Information Processing Systems. 3700–3710.
  23. 23.Paul Resnick and Hal R Varian. 1997. Recommender systems. Commun. ACM 40, 3 (1997), 56–58.
  24. 24.Ruslan Salakhutdinov and Andriy Mnih. 2007. Probabilistic Matrix Factorization. In 21th Conference on Neural Information Processing Systems, Vol. 1. 2–1.
  25. 25.David I Shuman, Sunil K Narang, Pascal Frossard, Antonio Ortega, and Pierre Vandergheynst. 2013. The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains. IEEE Signal Processing Magazine 30, 3 (2013), 83–98.
  26. 26.Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research 15, 1 (2014), 1929–1958.
  27. 27.Jiliang Tang, Charu Aggarwal, and Huan Liu. 2016. Recommendations in signed social networks. In Proceedings of the 25th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 31–40.
  28. 28.Jiliang Tang, Xia Hu, Huiji Gao, and Huan Liu. 2013. Exploiting local and global social context for recommendation.. In IJCAI, Vol. 13. 2712–2718.
  29. 29.Jiliang Tang, Xia Hu, and Huan Liu. 2013. Social recommendation: a review. Social Network Analysis and Mining 3, 4 (2013), 1113–1133.
  30. 30.Jiliang Tang, Suhang Wang, Xia Hu, Dawei Yin, Yingzhou Bi, Yi Chang, and Huan Liu. 2016. Recommendation with Social Dimensions. In AAAI. 251–257.
  31. 31.Tijmen Tieleman and Geoffrey Hinton. 2012. Lecture 6.5-RMSProp, COURSERA: Neural networks for machine learning. University of Toronto, Technical Report (2012).
  32. 32.Aaron Van den Oord, Sander Dieleman, and Benjamin Schrauwen. 2013. Deep content-based music recommendation. In Advances in neural Information Processing Systems. 2643–2651.
  33. 33.Hao Wang, Naiyan Wang, and Dit-Yan Yeung. 2015. Collaborative deep learning for recommender systems. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 1235–1244.
  34. 34.Suhang Wang, Jiliang Tang, Yilin Wang, and Huan Liu. 2018. Exploring Hierarchical Structures for Recommender Systems. IEEE Transactions on Knowledge and Data Engineering 30, 6 (2018), 1022–1035.
  35. 35.Xiang Wang, Xiangnan He, Liqiang Nie, and Tat-Seng Chua. 2017. Item silk road: Recommending items from information domains to social users. In Proceedings of the 40th International ACM SIGIR conference on Research and Development in Information Retrieval. ACM, 185–194.
  36. 36.Rongjing Xiang, Jennifer Neville, and Monica Rogati. 2010. Modeling relationship strength in online social networks. In Proceedings of the 19th international conference on World wide web. ACM, 981–990.
  37. 37.Bo Yang, Yu Lei, Jiming Liu, and Wenjie Li. 2017. Social collaborative filtering by trust. IEEE transactions on pattern analysis and machine intelligence 39, 8 (2017), 1633–1647.
  38. 38.Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical attention networks for document classification. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 1480–1489.
  39. 39.Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, and Jure Leskovec. 2018. Graph Convolutional Neural Networks for Web-Scale Recommender Systems. In KDD ’18. ACM, 974–983.
  40. 40.Lili Zhao, Zhongqi Lu, Sinno Jialin Pan, and Qiang Yang. 2016. Matrix Factorization+ for Movie Recommendation. In IJCAI. 3945–3951.
  41. 41.Xiangyu Zhao, Liang Zhang, Zhuoye Ding, Long Xia, Jiliang Tang, and Dawei Yin. 2018. Recommendations with Negative Feedback via Pairwise Deep Reinforcement Learning. In KDD’18. ACM, 1040–1048.
  42. 42.Zhou Zhao, Qifan Yang, Hanqing Lu, Tim Weninger, Deng Cai, Xiaofei He, and Yueting Zhuang. 2018. Social-aware movie recommendation via multimodal network learning. IEEE Transactions on Multimedia 20, 2 (2018), 430–440.

Citation

MLA
Fan, W., et al. “Graph Neural Networks for Social Recommendation”. The World Wide Web Conference, 2019, pp. 417–26, https://doi.org/10.1145/3308558.3313488.
APA
Fan, W., Ma, Y., Li, Q., He, Y., Zhao, E., Tang, J., & Yin, D. (2019). Graph Neural Networks for Social Recommendation. The World Wide Web Conference, 417–426. https://doi.org/10.1145/3308558.3313488
Chicago
Fan, W., Y. Ma, Q. Li, et al. 2019. “Graph Neural Networks for Social Recommendation”. The World Wide Web Conference, 417–26. https://doi.org/10.1145/3308558.3313488.
Harvard
Fan, W. et al. (2019) “Graph Neural Networks for Social Recommendation”, The World Wide Web Conference. ACM, pp. 417–426. Available at: https://doi.org/10.1145/3308558.3313488.
Vancouver
1. Fan W, Ma Y, Li Q, He Y, Zhao E, Tang J, Yin D (2019) Graph Neural Networks for Social Recommendation. In: The World Wide Web Conference. ACM, pp 417–426

BibTeX

@inproceedings{Fan_2019, series={WWW ’19}, title={Graph Neural Networks for Social Recommendation}, url={http://dx.doi.org/10.1145/3308558.3313488}, DOI={10.1145/3308558.3313488}, booktitle={The World Wide Web Conference}, publisher={ACM}, author={Fan, Wenqi and Ma, Yao and Li, Qing and He, Yuan and Zhao, Eric and Tang, Jiliang and Yin, Dawei}, year={2019}, month=May, pages={417–426}, collection={WWW ’19} }
Metadata:Crossref

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF