Content Provider Dynamics and Coordination in Recommendation Ecosystems

Omer Ben-Porat
Omer Ben-Porat
Itay Rosenberg
Itay Rosenberg

NeurIPS, 2020.

Cited by: 0|Bibtex|Views10
EI
Other Links: dblp.uni-trier.de|academic.microsoft.com
Weibo:
To support our vision of content provider coordination in RSs even further, we show in the appendix that the ratio between the social welfare of the best equilibrium and the worst equilibrium is unbounded

Abstract:

Recommendation Systems like YouTube are vibrant ecosystems with two types of users: Content consumers (those who watch videos) and content providers (those who create videos). While the computational task of recommending relevant content is largely solved, designing a system that guarantees high social welfare for all stakeholders is stil...More

Code:

Data:

0
Introduction
  • Recommendation systems (RSs hereinafter) play a major role in the life nowadays. Many modern RSs, like YouTube, Medium, or Spotify, recommend content created by others and go far beyond recommendations.
  • These fluctuations change the utility of the providers and the social welfare of the consumers, defined as the quality of their proposed content.
  • Given the demand for topic k, a strategy profile a, and the quality Q of the blogs for the selected topics in a, the recommendation function R recommends content, possibly in a randomized manner.
Highlights
  • Recommendation systems (RSs hereinafter) play a major role in our life nowadays
  • Successful content providers rely on the RS for some part of their income: Advertising, affiliated marketing, sponsorship, and merchandise; unsatisfied content providers might decide to provide a different type of content or even abandon the RS
  • We show that the provider dynamics always converges to a stable point, but the convergence time may be long
  • We suggest that a “you wrote this, would you be interested in writing on that?” feature could be revolutionary as well—contributing to better social welfare of content consumers, as well as the utility of content providers. Such a policy could be implemented in practice by a direct recommendation to providers, or by a more moderate action like nudging content providers to experiment with a different set of contents
  • To support our vision of content provider coordination in RSs even further, we show in the appendix that the ratio between the social welfare of the best equilibrium and the worst equilibrium is unbounded
  • Such a coordination between content providers may lead to a significant lift in social welfare
Results
  • The authors represent a game as a tuple P, T , D, Q, C, R, U , where P is the authors, T is the topics, D is the demand for topics, Q and C are the quality and conversion matrices, R is the recommendation function, and U is the utility function.
  • Let Hk(a) denote the set of authors whose documents have the highest quality among those who write on topic k under
  • For the purpose of social welfare maximization, it suffices to consider candidate profiles in which every topic is selected by at most one author.
  • If the authors create several copies of the same topic, high-quality players would block low-quality authors matched to it.
  • Tis the set of unmatched topics; Lk is a lower bound on the load on topic k, namely the ongoing number of players the authors matched to it; X, Y and E are the elements of the bipartite graph G (Y stores the set of unmatched players); and a∗ is a non-valid, empty profile that the authors construct as the algorithm advances.
  • The authors first find the set of highest-quality players for every topic k, denoted Ak (Line 7).
  • For every topic k, the authors consider the set of most profitable players w.r.t. k and their potential utility if matched to k.
  • In Line 16 the authors use M to set the strategies of the players in NG(W ): Every player j ∈ NG(W ) is matched to the topic associated with the node M (j) ∈ W .
Conclusion
  • Focus on the first time a non-empty saturated set W was found in Line 13, and denote the iteration index by t .
  • The following Lemma 1 shows that as long as a bipartite G satisfies Hall’s marriage condition, the authors can find the maximum saturated set W efficiently.
  • The authors suggest that a “you wrote this, would you be interested in writing on that?” feature could be revolutionary as well—contributing to better social welfare of content consumers, as well as the utility of content providers.
Summary
  • Recommendation systems (RSs hereinafter) play a major role in the life nowadays. Many modern RSs, like YouTube, Medium, or Spotify, recommend content created by others and go far beyond recommendations.
  • These fluctuations change the utility of the providers and the social welfare of the consumers, defined as the quality of their proposed content.
  • Given the demand for topic k, a strategy profile a, and the quality Q of the blogs for the selected topics in a, the recommendation function R recommends content, possibly in a randomized manner.
  • The authors represent a game as a tuple P, T , D, Q, C, R, U , where P is the authors, T is the topics, D is the demand for topics, Q and C are the quality and conversion matrices, R is the recommendation function, and U is the utility function.
  • Let Hk(a) denote the set of authors whose documents have the highest quality among those who write on topic k under
  • For the purpose of social welfare maximization, it suffices to consider candidate profiles in which every topic is selected by at most one author.
  • If the authors create several copies of the same topic, high-quality players would block low-quality authors matched to it.
  • Tis the set of unmatched topics; Lk is a lower bound on the load on topic k, namely the ongoing number of players the authors matched to it; X, Y and E are the elements of the bipartite graph G (Y stores the set of unmatched players); and a∗ is a non-valid, empty profile that the authors construct as the algorithm advances.
  • The authors first find the set of highest-quality players for every topic k, denoted Ak (Line 7).
  • For every topic k, the authors consider the set of most profitable players w.r.t. k and their potential utility if matched to k.
  • In Line 16 the authors use M to set the strategies of the players in NG(W ): Every player j ∈ NG(W ) is matched to the topic associated with the node M (j) ∈ W .
  • Focus on the first time a non-empty saturated set W was found in Line 13, and denote the iteration index by t .
  • The following Lemma 1 shows that as long as a bipartite G satisfies Hall’s marriage condition, the authors can find the maximum saturated set W efficiently.
  • The authors suggest that a “you wrote this, would you be interested in writing on that?” feature could be revolutionary as well—contributing to better social welfare of content consumers, as well as the utility of content providers.
Related work
  • Strikingly, content provider welfare and their fair treatment were only suggested very recently in the Recommendation Systems and Information Retrieval communities [12, 14, 18, 35, 40, 46]. All of these works do not model the incentives of content providers explicitly, and consequently cannot offer a what-if analysis like ours.

    Our model is similar to those employed in several recent papers [4, 5, 7, 8, 30]. Ben-Porat et al [8] study a model that is a special case of ours, and show that every learning dynamic converges. Our Theorem 1 recovers and extends their convergence results. Moreover, unlike this work, they do not address convergence time, social welfare, and centralized equilibrium computation. Other works [5, 7, 30] aim to design recommendation mechanisms that mitigate strategic behavior and lead to long-term welfare. On the negative side, their mechanisms might knowingly recommend inferior content to some consumers. We see their work as parallel to ours, as in this work we focus on the prevailing recommendation approach—recommending the best-fitting content. We suggest that a centralized approach, in which the RS orchestrates the player-topic matching, can significantly improve the time until the system reaches stability (in the form of equilibrium). Furthermore, we envision that our approach can also lead to high social welfare, as we discuss in Section 5.
Funding
  • Ben-Porat is partially funded by a PhD fellowship from JPMorgan Chase & Co
  • Tennenholtz is funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement n◦ 740435)
Reference
  • Y. C. Academy. Lesson: Find your niche. URL https://creatoracademy.youtube.com/page/lesson/niche.
    Findings
  • E. Anshelevich, A. Dasgupta, J. Kleinberg, E. Tardos, T. Wexler, and T. Roughgarden. The price of stability for network design with fair cost allocation. SIAM Journal on Computing, 38(4):1602–1623, 2008.
    Google ScholarLocate open access versionFindings
  • R. Baeza-Yates, B. Ribeiro-Neto, et al. Modern information retrieval, volume 46ACM press New York, 1999.
    Google ScholarLocate open access versionFindings
  • R. Ben Basat, M. Tennenholtz, and O. Kurland. A game theoretic analysis of the adversarial retrieval setting. Journal of Artificial Intelligence Research, 60:1127–1164, 2017.
    Google ScholarLocate open access versionFindings
  • O. Ben-Porat and M. Tennenholtz. A game-theoretic approach to recommendation systems with strategic content providers. In Advances in Neural Information Processing Systems (NIPS) 2018, 2018.
    Google ScholarLocate open access versionFindings
  • O. Ben-Porat and M. Tennenholtz. Regression equilibrium. In Proceedings of the 2019 ACM Conference on Economics and Computation, pages 173–191, 2019.
    Google ScholarLocate open access versionFindings
  • O. Ben-Porat, G. Goren, I. Rosenberg, and M. Tennenholtz. From recommendation systems to facility location games. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 1772–1779, 2019.
    Google ScholarLocate open access versionFindings
  • O. Ben-Porat, I. Rosenberg, and M. Tennenholtz. Convergence of learning dynamics in information retrieval games. In The Thirty-Third AAAI Conference on Artificial Intelligence (AAAI 2019), 2019.
    Google ScholarLocate open access versionFindings
  • J. Bennett and S. Lanning. The netflix prize. In Proceedings of KDD cup and workshop, volume 2007, page 35, 2007.
    Google ScholarLocate open access versionFindings
  • O. Besbes, Y. Gur, and A. Zeevi. Optimization in online content recommendation services: Beyond click-through rates. Manufacturing & Service Operations Management, 18(1):15–33, 2016.
    Google ScholarLocate open access versionFindings
  • A. Beutel, J. Chen, T. Doshi, H. Qian, L. Wei, Y. Wu, L. Heldt, Z. Zhao, L. Hong, E. H. Chi, et al. Fairness in recommendation ranking through pairwise comparisons. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2212–2220, 2019.
    Google ScholarLocate open access versionFindings
  • R. Burke. Multisided fairness for recommendation. arXiv preprint arXiv:1707.00093, 2017.
    Findings
  • O. Butman, A. Shtok, O. Kurland, and D. Carmel. Query-performance prediction using minimal relevance feedback. In Proceedings of ICTIR, page 7, 2013.
    Google ScholarLocate open access versionFindings
  • C. Castillo. Fairness and transparency in ranking. In ACM SIGIR Forum, volume 52, pages 64–71. ACM, 2019.
    Google ScholarLocate open access versionFindings
  • F. Chierichetti, R. Kumar, S. Lattanzi, and S. Vassilvitskii. Fair clustering through fairlets. In Advances in Neural Information Processing Systems, pages 5036–5044, 2017.
    Google ScholarLocate open access versionFindings
  • J. Cohen, A. Héliou, and P. Mertikopoulos. Learning with bandit feedback in potential games. In Proceedings of the 31th International Conference on Neural Information Processing Systems, 2017.
    Google ScholarLocate open access versionFindings
  • C. Dwork, M. Hardt, T. Pitassi, O. Reingold, and R. Zemel. Fairness through awareness. In Proceedings of the 3rd innovations in theoretical computer science conference, pages 214–226. ACM, 2012.
    Google ScholarLocate open access versionFindings
  • M. D. Ekstrand, R. Burke, and F. Diaz. Fairness and discrimination in retrieval and recommendation. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1403–1404. ACM, 2019.
    Google ScholarLocate open access versionFindings
  • E. Even-Dar, A. Kesselman, and Y. Mansour. Convergence time to nash equilibrium in load balancing. ACM Transactions on Algorithms (TALG), 3(3):32–es, 2007.
    Google ScholarLocate open access versionFindings
  • D. Gale and L. S. Shapley. College admissions and the stability of marriage. The American Mathematical Monthly, 69(1):9–15, 1962.
    Google ScholarLocate open access versionFindings
  • V. Garg and T. Jaakkola. Learning tree structured potential games. In Advances In Neural Information Processing Systems (NIPS), pages 1552–1560, 2016.
    Google ScholarLocate open access versionFindings
  • A. Ghose, A. Goldfarb, and S. P. Han. How is the mobile internet different? search costs and local activities. Information Systems Research, 24(3):613–631, 2012.
    Google ScholarLocate open access versionFindings
  • E. Ie, V. Jain, J. Wang, S. Narvekar, R. Agarwal, R. Wu, H.-T. Cheng, T. Chandra, and C. Boutilier. Slateq: A tractable decomposition for reinforcement learning with recommendation sets. In Proceedings of the Twenty-eighth International Joint Conference on Artificial Intelligence (IJCAI-19), pages 2592–2599, Macau, China, 2019.
    Google ScholarLocate open access versionFindings
  • T. Joachims, L. A. Granka, B. Pan, H. Hembrooke, and G. Gay. Accurately interpreting clickthrough data as implicit feedback. In Proceedings of SIGIR, pages 154–161, 2005.
    Google ScholarLocate open access versionFindings
  • Y. Koren, R. Bell, C. Volinsky, et al. Matrix factorization techniques for recommender systems. Computer, 42(8):30–37, 2009.
    Google ScholarLocate open access versionFindings
  • J. Leonhardt, A. Anand, and M. Khosla. User fairness in recommender systems. In Companion Proceedings of The Web Conference 2018, pages 101–102, 2018.
    Google ScholarLocate open access versionFindings
  • C. Liu and Y. Wei. The impacts of time constraint on users’ search strategy during search process. Proceedings of the Association for Information Science and Technology, 53(1), 2016.
    Google ScholarLocate open access versionFindings
  • L. Liu, S. Dean, E. Rolf, M. Simchowitz, and M. Hardt. Delayed impact of fair machine learning. In International Conference on Machine Learning, pages 3156–3164, 2018.
    Google ScholarLocate open access versionFindings
  • R. Mehrotra, J. McInerney, H. Bouchard, M. Lalmas, and F. Diaz. Towards a fair marketplace: Counterfactual evaluation of the trade-off between relevance, fairness & satisfaction in recommendation systems. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management, pages 2243–2251. ACM, 2018.
    Google ScholarLocate open access versionFindings
  • M. Mladenov, E. Creager, O. Ben-Porat, K. Swersky, R. Zemel, and C. Boutilier. Optimizing long-term social welfare in recommender systems: A constrained matching approach. 2020.
    Google ScholarFindings
  • N. Modani, D. Jain, U. Soni, G. K. Gupta, and P. Agarwal. Fairness aware recommendations on behance. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pages 144–155.
    Google ScholarLocate open access versionFindings
  • D. Monderer and L. S. Shapley. Potential games. Games and economic behavior, 14(1):124–143, 1996.
    Google ScholarLocate open access versionFindings
  • N. Nisan, T. Roughgarden, E. Tardos, and V. V. Vazirani. Algorithmic game theory, volume 1. Cambridge University Press Cambridge, 2007.
    Google ScholarFindings
  • G. Palaiopanos, I. Panageas, and G. Piliouras. Multiplicative weights update with constant step-size in congestion games: Convergence, limit cycles and chaos. In Advances in Neural Information Processing Systems (NIPS), pages 5874–5884, 2017.
    Google ScholarLocate open access versionFindings
  • G. K. Patro, A. Biswas, N. Ganguly, K. P. Gummadi, and A. Chakraborty. Fairrec: Two-sided fairness for personalized recommendations in two-sided platforms. In Proceedings of The Web Conference 2020, WWW ’20, page 1194–1204, New York, NY, USA, 2020. Association for Computing Machinery. ISBN 9781450370233. doi: 10.1145/3366423.3380196.
    Locate open access versionFindings
  • D. Pedreshi, S. Ruggieri, and F. Turini. Discrimination-aware data mining. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 560–568. ACM, 2008.
    Google ScholarLocate open access versionFindings
  • L. Pizzato, T. Rej, T. Chung, I. Koprinska, and J. Kay. Recon: a reciprocal recommender for online dating. In Proceedings of the Fourth ACM Conference on Recommender Systems, pages 207–214. ACM, 2010.
    Google ScholarLocate open access versionFindings
  • G. Pleiss, M. Raghavan, F. Wu, J. Kleinberg, and K. Q. Weinberger. On fairness and calibration. In Advances in Neural Information Processing Systems, pages 5684–5693, 2017.
    Google ScholarLocate open access versionFindings
  • S. E. Robertson. The probability ranking principle in ir. Journal of documentation, 33(4):294–304, 1977.
    Google ScholarLocate open access versionFindings
  • A. Singh and T. Joachims. Fairness of exposure in rankings. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2219–2228. ACM, 2018.
    Google ScholarLocate open access versionFindings
  • T. Sühr, A. J. Biega, M. Zehlike, K. P. Gummadi, and A. Chakraborty. Two-sided fairness for repeated matchings in two-sided markets: A case study of a ride-hailing platform. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 3082–3092. ACM, 2019.
    Google ScholarLocate open access versionFindings
  • M. Tennenholtz and O. Kurland. Rethinking search engines and recommendation systems: a game theoretic perspective. Communications of the ACM, 62(12):66–75, 2019.
    Google ScholarLocate open access versionFindings
  • S. Yao and B. Huang. Beyond parity: Fairness objectives for collaborative filtering. In Advances in Neural Information Processing Systems, pages 2921–2930, 2017.
    Google ScholarLocate open access versionFindings
  • H. Yu, C. Liu, and F. ZHANG. Reciprocal recommendation algorithm for the field of recruitment. Journal of Information & Computational Science, 8(16):4061–4068, 2011.
    Google ScholarLocate open access versionFindings
  • R. Zemel, Y. Wu, K. Swersky, T. Pitassi, and C. Dwork. Learning fair representations. In Proceedings of the 30th International Conference on Machine Learning (ICML-13), pages 325–333, 2013.
    Google ScholarLocate open access versionFindings
  • Y. Zheng. Multi-stakeholder recommendation: Applications and challenges. arXiv preprint arXiv:1707.08913, 2017.
    Findings
Full Text
Your rating :
0

 

Tags
Comments