AI帮你理解科学

AI 生成解读视频

AI抽取解析论文重点内容自动生成视频


pub
查看解读视频

AI 溯源

AI解析本论文相关学术脉络


Master Reading Tree
生成 溯源树

AI 精读

AI抽取本论文的概要总结


微博一下
We further study to what extent users’ “wow” and click behavior can be predicted from their social connections

Understanding WeChat User Preferences and "Wow" Diffusion

IEEE Transactions on Knowledge and Data Engineering, pp.1-1, (2021)

被引用0|浏览42596
下载 PDF 全文
引用
微博一下

摘要

WeChat is the largest social instant messaging platform in China with 1.1 billion monthly active users.“Top Stories“ is a kind of novel friend-enhanced recommendation engine in WeChat, in which users can read articles based on both their own and their friends\u0027 preferences. Specifically, when a user reads an article by opening it, the...更多
0
简介
  • The author NFORMATIONs diffusion [33] has increasingly changed from offline to online these years.
  • Despite popular applications and extensive studies of information diffusion algorithms, it is still unclear about the inherent factors that result in different types of user feedback.
  • Haoran Wang, Zihan, Aliya, Victoria, Sunny, Shine, Wei Zhang.
  • These digital libraries are free and open now!
  • The authors aim at understanding users’ “wow” and click behavior from different aspects, including user demographics, dyadic and triadic correlations, and users’ ego network structures
重点内容
  • I NFORMATION diffusion [33] has increasingly changed from offline to online these years
  • Our experiments show that the proposed solution can consistently outperform alternative methods
  • The comparison methods can be roughly divided into several categories: (1) traditional classifiers: Logistic Regression (LR) and Random Forest (RF), (2) deep learning method by modeling feature interactions: xDeepFM, (3) the state-of-the-art user behavior prediction methods based on ego networks: DeepInf and Wang et al and (4) hierarchical graph representation learning methods: SAGPool, ASAP and StructPool. (3) and (4) are both GNNbased methods
  • We observe that our model DiffuseGNN consistently outperforms baseline methods. It can not achieve better prediction performance than other methods, it leverages hand-craft user features, correlation features, and network features. xDeepFM, a factorization-machine based neural network model, achieves better performance than LR and RF on WeChat dataset, which might imply that the correlation between user features is an inherent factor that impacts users’ “wow” and click behaviors, such as the correlation between users’ gender and age
  • The prediction performance of Wang et al is inferior to DeepInf, which might indicate that sometimes local topological features could result in a negative impact on user behavior prediction
  • Our experiments show that the proposed method can significantly improve the prediction performance compared with alternative methods
  • 3) Given the fixed number of friends who “wow”ed an article, the larger #Connected Components (CC), the lower the “wow” probability of ego users is, but the higher the click probability is. Based on these important discoveries, we develop a unified model DiffuseGNN to predict users’ behaviors. We evaluate it on the real sizable social networks, and results show that the proposed model can achieve significantly better performance over several state-of-the-arts
方法
  • The authors compare the proposed model with the following methods.

    Random. The authors generate like/click probability uniformly in the range [0, 1) for prediction.

    Logistic Regression (LR).
结果
  • The authors observe that the model DiffuseGNN consistently outperforms baseline methods.
  • For traditional classifiers, it can not achieve better prediction performance than other methods, it leverages hand-craft user features, correlation features, and network features.
  • XDeepFM, a factorization-machine based neural network model, achieves better performance than LR and RF on WeChat dataset, which might imply that the correlation between user features is an inherent factor that impacts users’ “wow” and click behaviors, such as the correlation between users’ gender and age.
  • The prediction performance of Wang et al is inferior to DeepInf, which might indicate that sometimes local topological features could result in a negative impact on user behavior prediction
结论
  • When users’ click and “wow” behaviors are concerned, it is natural to take into account the article content.
  • 3) Given the fixed number of friends who “wow”ed an article, the larger #CC, the lower the “wow” probability of ego users is, but the higher the click probability is.
  • Based on these important discoveries, the authors develop a unified model DiffuseGNN to predict users’ behaviors.
  • The authors evaluate it on the real sizable social networks, and results show that the proposed model can achieve significantly better performance over several state-of-the-arts
表格
  • Table1: User activity w.r.t. gender
  • Table2: Dyadic correlations w.r.t. user gender and friend gender
  • Table3: Dyadic correlations w.r.t. the distance between the user and the friend
  • Table4: Dyadic correlations w.r.t. users’ and friends’ social roles. OU: ordinary user; OL: opinion leader
  • Table5: Dyadic correlations w.r.t. users’ and friends’ social roles. OU: ordinary user; SH: structural hole
  • Table6: List of input features of DiffuseGNN. (*x) indicates the dimension of the feature. OL: opinion leader. SH: structural hole
  • Table7: Results of User Behavior Prediction
Download tables as Excel
相关工作
  • 6.1 Social Influence Analysis

    Social influence has been studied and modeled widely from different viewpoints. At the macro level, the problem of influence maximization in social networks has been studied in [6], [17]. Xin et al [35] study the indirect influence on Twitter. Micro influence, like pairwise influence, has been studied in [12], [34], [50]. Liu et al [24] study the micro mechanism of influence diffusion in heterogeneous social networks and propose a probabilistic generative model. Tang et al [38] propose Topical Affinity Propagation (TAP) to model influence on different topics in the academic network [39] and the heterogeneous network. More recently, deep learning models have been proposed to model social influence. Qiu et al [31] use Graph Attention Networks (GAT) to model social influence locality. Feng et al [10] propose a skip-gram architecture to learn user embeddings that reflect social influence. In this work, we first analyze microlevel social influence (e.g., dyadic and triadic correlations) and local network structure, based on which we propose an effective method to model social influence.
基金
  • This work is supported by a research fund of Tsinghua- Tencent Joint Laboratory for Internet Innovation Technology, the National Key R&D Program of China (2018YFB1402600), NSFC for Distinguished Young Scholar (61825602), NSFC (61836013), NSFC (61672313), and NSF under grants III-1763325, III-1909323, and SaTC-1930941
研究对象与分析
users: 48084772
To avoid overfitting, we further select a subset of the data by first extracting users who performed at least ten interactions (“wow” or click), and then extracting these users’ friendship networks and their attributes. The final dataset contains 48,084,772 users, 61,252,317 articles, and 7,459,660,092 interactions. All data are preprocessed via data masking to protect user privacy

datasets: 3
For traditional classifiers, it can not achieve better prediction performance than other methods, although it leverages hand-craft user features, correlation features, and network features. xDeepFM, a factorization-machine based neural network model, achieves better performance than LR and RF on WeChat dataset, which might imply that the correlation between user features is an inherent factor that impacts users’ “wow” and click behaviors, such as the correlation between users’ gender and age. DeepInf and Wang et al could both achieve good prediction performance on three datasets. It demonstrates that modeling dyadic correlations via graph attention is effective

引用论文
  • L. C. Andrews. Special functions of mathematics for engineers, volume 49. Spie Press, 1998.
    Google ScholarFindings
  • R. S. Burt. Structural holes and good ideas. American journal of sociology, 110(2):349–399, 2004.
    Google ScholarLocate open access versionFindings
  • S. Cao, W. Lu, and Q. Xu. Deep neural networks for learning graph representations. In AAAI’16.
    Google ScholarFindings
  • S. Cao, W. Lu, and Q. Xu. Grarep: Learning graph representations with global structural information. In CIKM’15, pages 891–900.
    Google ScholarLocate open access versionFindings
  • H. Chen, B. Perozzi, Y. Hu, and S. Skiena. Harp: Hierarchical representation learning for networks. In AAAI’18.
    Google ScholarFindings
  • P. Domingos and M. Richardson. Mining the network value of customers. In KDD’01, page 57–66.
    Google ScholarLocate open access versionFindings
  • B. L. Douglas. The weisfeiler-lehman method and graph isomorphism testing. arXiv preprint arXiv:1101.5211, 2011.
    Findings
  • J. Duchi, E. Hazan, and Y. Singer. Adaptive subgradient methods for online learning and stochastic optimization. JMLR, 12:2121– 2159, 2011.
    Google ScholarLocate open access versionFindings
  • L. Feng, Y. Hu, B. Li, H. E. Stanley, S. Havlin, and L. A. Braunstein. Competing for attention in social media under information overload conditions. PloS one, 10(7), 2015.
    Google ScholarLocate open access versionFindings
  • S. Feng, G. Cong, A. Khan, X. Li, Y. Liu, and Y. M. Chee. Inf2vec: Latent representation model for social influence embedding. In ICDE’18, pages 941–952.
    Google ScholarLocate open access versionFindings
  • J. Goldenberg, B. Libai, and E. Muller. Talk of the network: A complex systems look at the underlying process of word-ofmouth. Marketing letters, 12(3):211–223, 2001.
    Google ScholarLocate open access versionFindings
  • A. Goyal, F. Bonchi, and L. V. Lakshmanan. Learning influence probabilities in social networks. In WSDM’10, page 241–250.
    Google ScholarLocate open access versionFindings
  • M. Granovetter. Threshold models of collective behavior. American journal of sociology, 83(6):1420–1443, 1978.
    Google ScholarLocate open access versionFindings
  • T. H. Haveliwala. Topic-sensitive pagerank: A context-sensitive ranking algorithm for web search. IEEE TKDE, 15(4):784–796, 2003.
    Google ScholarLocate open access versionFindings
  • G. Jawaheer, M. Szomszor, and P. Kostkova. Comparison of implicit and explicit feedback from an online music recommendation service. In Proceedings of the 1st international workshop on information heterogeneity and fusion in recommender systems, pages 47–51, 2010.
    Google ScholarLocate open access versionFindings
  • G. Jawaheer, P. Weller, and P. Kostkova. Modeling user preferences in recommender systems: A classification framework for explicit and implicit user feedback. ACM Transactions on Interactive Intelligent Systems (TiiS), 4(2):1–26, 2014.
    Google ScholarLocate open access versionFindings
  • D. Kempe, J. Kleinberg, and E. Tardos. Maximizing the spread of influence through a social network. In KDD’03, page 137–146.
    Google ScholarLocate open access versionFindings
  • T. N. Kipf and M. Welling. Semi-supervised classification with graph convolutional networks. In ICLR’17.
    Google ScholarLocate open access versionFindings
  • J. Lee, I. Lee, and J. Kang. Self-attention graph pooling. In 36th International Conference on Machine Learning, ICML 2019, pages 6661–6670. International Machine Learning Society (IMLS), 2019.
    Google ScholarLocate open access versionFindings
  • R. Li, S. Wang, F. Zhu, and J. Huang. Adaptive graph convolutional neural networks. In AAAI’18.
    Google ScholarFindings
  • J. Lian, X. Zhou, F. Zhang, Z. Chen, X. Xie, and G. Sun. xdeepfm: Combining explicit and implicit feature interactions for recommender systems. In KDD’18, pages 1754–1763.
    Google ScholarLocate open access versionFindings
  • A. Liaw, M. Wiener, et al. Classification and regression by randomforest. R news, 2(3):18–22, 2002.
    Google ScholarLocate open access versionFindings
  • H. Lin, J. Jia, J. Qiu, Y. Zhang, G. Shen, L. Xie, J. Tang, L. Feng, and T.-S. Chua. Detecting stress based on social interactions in social networks. IEEE TKDE, 29(9):1820–1833, 2017.
    Google ScholarLocate open access versionFindings
  • L. Liu, J. Tang, J. Han, and S. Yang. Learning influence from heterogeneous social networks. Data Mining and Knowledge Discovery, 25, 11 2012.
    Google ScholarLocate open access versionFindings
  • N. N. Liu, E. W. Xiang, M. Zhao, and Q. Yang. Unifying explicit and implicit feedback for collaborative filtering. In CIKM’10, pages 1445–1448.
    Google ScholarLocate open access versionFindings
  • Y. Matsubara, Y. Sakurai, B. A. Prakash, L. Li, and C. Faloutsos. Rise and fall patterns of information diffusion: model and implications. In KDD’12, pages 6–14.
    Google ScholarLocate open access versionFindings
  • M. Niepert, M. Ahmed, and K. Kutzkov. Learning convolutional neural networks for graphs. In ICML’16, pages 2014–2023.
    Google ScholarLocate open access versionFindings
  • L. Page, S. Brin, R. Motwani, and T. Winograd. The pagerank citation ranking: Bringing order to the web. Technical report, Stanford InfoLab, 1999.
    Google ScholarFindings
  • B. Perozzi, R. Al-Rfou, and S. Skiena. Deepwalk: Online learning of social representations. In KDD’14, pages 701–710.
    Google ScholarLocate open access versionFindings
  • J. Qiu, Y. Dong, H. Ma, J. Li, K. Wang, and J. Tang. Network embedding as matrix factorization: Unifying deepwalk, line, pte, and node2vec. In WSDM’18, pages 459–467.
    Google ScholarLocate open access versionFindings
  • J. Qiu, J. Tang, H. Ma, Y. Dong, K. Wang, and J. Tang. Deepinf: Social influence prediction with deep learning. In KDD’18, pages 2110–2119.
    Google ScholarLocate open access versionFindings
  • E. Ranjan, S. Sanyal, and P. P. Talukdar. Asap: Adaptive structure aware pooling for learning hierarchical graph representations. In AAAI, pages 5470–5477, 2020.
    Google ScholarLocate open access versionFindings
  • E. M. Rogers. Diffusion of innovations. Simon and Schuster, 2010.
    Google ScholarLocate open access versionFindings
  • K. Saito, R. Nakano, and M. Kimura. Prediction of information diffusion probabilities for independent cascade model. In KES’08, page 67–75.
    Google ScholarLocate open access versionFindings
  • X. Shuai, Y. Ding, J. Busemeyer, S. Chen, Y. Sun, and J. Tang. Modeling indirect influence on twitter. International Journal on Semantic Web and Information Systems (IJSWIS), 8(4):20–36, 2012.
    Google ScholarLocate open access versionFindings
  • F.-Y. Sun, J. Hoffmann, and J. Tang. Infograph: Unsupervised and semi-supervised graph-level representation learning via mutual information maximization. In ICLR’20.
    Google ScholarFindings
  • J. Tang, M. Qu, M. Wang, M. Zhang, J. Yan, and Q. Mei. Line: Large-scale information network embedding. In WWW’15, pages 1067–1077.
    Google ScholarLocate open access versionFindings
  • J. Tang, J. Sun, C. Wang, and Z. Yang. Social influence analysis in large-scale networks. In KDD’09, pages 807–816.
    Google ScholarLocate open access versionFindings
  • J. Tang, J. Zhang, L. Yao, J. Li, L. Zhang, and Z. Su. Arnetminer: extraction and mining of academic social networks. In KDD’08, pages 990–998.
    Google ScholarLocate open access versionFindings
  • L. Tang, B. Long, B.-C. Chen, and D. Agarwal. An empirical study on recommendation with multiple types of feedback. In KDD’16, pages 283–292.
    Google ScholarLocate open access versionFindings
  • R. Tarjan. Depth-first search and linear graph algorithms. SIAM journal on computing, 1(2):146–160, 1972.
    Google ScholarLocate open access versionFindings
  • J. Ugander, L. Backstrom, C. Marlow, and J. Kleinberg. Structural diversity in social contagion. PNAS, 109(16):5962–5966, 2012.
    Google ScholarLocate open access versionFindings
  • P. Velickovic, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio. Graph attention networks. In ICLR’18.
    Google ScholarLocate open access versionFindings
  • H. Wang, Q. Meng, J. Fan, Y. Li, L. Cui, X. Zhao, C. Peng, G. Chen, and X. Du. Social influence does matter: User action prediction for in-feed advertising. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 246–253, 2020.
    Google ScholarLocate open access versionFindings
  • K. Xu, W. Hu, J. Leskovec, and S. Jegelka. How powerful are graph neural networks? In ICLR’19.
    Google ScholarLocate open access versionFindings
  • Y. Yang, J. Tang, C. Leung, Y. Sun, Q. Chen, J. Li, and Q. Yang. Rain: Social role-aware information diffusion. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 29, 2015.
    Google ScholarLocate open access versionFindings
  • Z. Ying, J. You, C. Morris, X. Ren, W. Hamilton, and J. Leskovec. Hierarchical graph representation learning with differentiable pooling. In NeurIPS’18, pages 4800–4810.
    Google ScholarLocate open access versionFindings
  • H. Yuan and S. Ji. Structpool: Structured graph pooling via conditional random fields. In International Conference on Learning Representations, 2020.
    Google ScholarLocate open access versionFindings
  • J. Zhang, Y. Dong, Y. Wang, J. Tang, and M. Ding. Prone: fast and scalable network representation learning. In IJCAI’19, pages 4278–4284.
    Google ScholarLocate open access versionFindings
  • J. Zhang, B. Liu, J. Tang, T. Chen, and J. Li. Social influence locality for modeling retweeting behaviors. In IJCAI’13, pages 2761–2767. Ruobing Xie is a researcher of WeChat, Tencent. He got his BEng degree in 2014 and his master degree in 2017 from the Department of Computer Science and Technology, Tsinghua University. His research interests are natural language processing and recommender system. He has published over 15 papers in international journals and conferences including IJCAI, AAAI, ACL and EMNLP.
    Google ScholarLocate open access versionFindings
  • Jie Tang is a Professor and the Associate Chair of the Department of Computer Science at Tsinghua University. He is a Fellow of the IEEE. His interests include artificial intelligence, data mining, social networks, and machine learning. He was honored with the SIGKDD Testof-Time Award, the UK Royal Society-Newton Advanced Fellowship Award, NSFC for Distinguished Young Scholar, and KDD’18 Service Award.
    Google ScholarFindings
  • Kai Zhuang received his master degree from the Department of Computer Science and Technology, Sichuan University, China, in 2013. He is a researcher in WeChat Search Application Department, Tencent Inc., China.
    Google ScholarLocate open access versionFindings
  • Xu Zhang received his master degree in the College of Computer Science and Engineering, Northeastern University in 2008. He is the team leader of WeChat User Profile, Tencent. His research interests include machine learning and its applications, such as social networks, user interest models and user behavior research.
    Google ScholarLocate open access versionFindings
  • Leyu Lin received his master degree from the Institute of Computing Technology, Chinese Academy of Sciences, in 2008. He is currently the deputy director of WeChat Search Team, Tencent. His research interests include machine learning and its applications, such as search system, recommendation systems and computational advertising.
    Google ScholarFindings
  • Yuxiao Dong received his Ph.D. from University of Notre Dame in 2017. He is a Senior Researcher at Microsoft Research Redmond. His research focuses on graph representation learning, knowledge graphs, social & information networks, and data mining, with an emphasis on developing machine learning models to address problems in Web-scale systems.
    Google ScholarFindings
  • Philip S. Yu received the PhD degree in electrical engineering from Stanford University. He is a distinguished professor in computer science with the University of Illinois, Chicago, and is the Wexler chair in Information Technology. His research interests include big data, data mining, data stream, database, and privacy. He was the editor-in-chief of the IEEE Transactions on Knowledge and Data Engineering and the ACM Transactions on Knowledge Discovery from Data. He received the ACM SIGKDD 2016 Innovation Award, a Research Contributions Award from the IEEE International Conference on Data Mining (2003), and a Technical Achievement Award from the IEEE Computer Society (2013). He is a fellow of the IEEE and ACM.
    Google ScholarLocate open access versionFindings
您的评分 :
0

 

标签
评论
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科