Elsevier

Information Sciences

Volume 542, 4 January 2021, Pages 1-23
Information Sciences

Sequential dynamic event recommendation in event-based social networks: An upper confidence bound approach

https://doi.org/10.1016/j.ins.2020.06.047Get rights and content

Abstract

In recent years, there have been some platforms that have focused on recommending commodities or events to users using event-based social networks (EBSNs). Some studies have attempted to find the optimal recommendation sequence of these items, assuming that the sequence stops once the user accepts one recommendation or the item list runs out. However, in reality, social media platforms will not stop recommending different commodities or social events to users until the user becomes bored and abandons the platform. Since it is 5 to 25 times more difficult to attract a new user than to retain an old one,1 it would be helpful if the platform could determine when to stop making recommendations. In this work, we investigate the problem of sequential dynamic event recommendation with feedback (SDERF), where the platform continues recommending events even when the user has accepted one that is satisfactory. We first model the SDERF problem and provide two variants, namely, an online learning model with/without contextual information. Then, we apply an upper confidence bound (UCB) approach with an expected regret polynomial in the number of events and rounds. Finally, we evaluate the performance of our proposed algorithms using both real and synthetic datasets.

Introduction

In recent years, event-based social networks (EBSNs) have become popular social media services. An EBSN mainly publishes events on a platform through organizers. Then, users register online and join some groups of events, and users participate in these events offline [1]. Examples include Meetup 2, Eventbrite 3 and Plancast 4. On the Meetup platform, users can create groups online and organize events offline through groups for other users to join. On the Plancast and Eventbrite platforms, events in a community are organized, and users and organizers socialize in that community [1], [2].

Unfortunately, most of the existing research mainly considers the events recommended by the platform to users and the default user introduction activities [2], [3], [4], [5], [6], [7]. In fact, users can refuse the activities recommended by the platform. For example, on the platform, we may find that Amy’s interest is music and that the platform recommended rock to her, although she prefers light music. In this case, Amy would refuse the rock music and leave the platform. In addition, if a user has faced this situation many times, s/he is likely to uninstall the EBSN platform. Therefore, to obtain better feedback from users, we should allow the events recommended by the platform to be accepted or rejected.

In particular, current studies have not considered the order of recommended events when recommending multiple events to users. Since the order of the recommended events is related to the content of the activities the user sees, if the user sees an event that s/he is not interested in, the user may be less upset, so s/he will not continue to see other events. For example, suppose that on a weekend, the platform recommends three activities for Bob: mountain climbing, a marathon and a concert. However, from Monday to Friday, Bob works overtime continuously. Therefore, he does not have the extra energy to participate in outdoor sports on weekends; he wishes to go to concerts more. If he first sees that the events are climbing and a marathon, he will choose to reject them and may uninstall the platform. Therefore, we must consider the recommendation order because it can better control the fatigue of users.

In addition, considering the revenue of the platform, an EBSN platform should continuously give users recommended events. For example, when a user logs on to the EBSN platform, the platform recommends several activities to the user according to the attributes of the user. The EBSN platform should continue to recommend events to the user when new events are launched so that the platform can retain some potential users who may continue to participate in activities. Therefore, considering so many complex factors, we urgently need a high-quality event-recommendation algorithm to ensure that users achieve higher satisfaction.

To consider user feedback and continuously recommend sequential events, we refer to this problem as sequential dynamic event recommendation with feedback (SDERF). We consider two variants of this problem, namely, an online learning model with no contextual information, in which all users share similar intrinsic interests. Furthermore, the probability of a user accepting an event is independent of other events in the recommended list. For example, when a user arrives at the platform, the user has a degree of interest in the event, so we can estimate some parameters. When the next user arrives at the platform, we can still use the parameters estimated by the previous user to calculate the user’s contribution to the platform (platform revenue). Moreover, in an online learning model with contextual information, different users have different levels of tolerance for unsatisfactory recommendations. Furthermore, the probability of a user accepting an event is not independent of other events in the recommended list. For example, when users arrive at the platform, they have a different degree of interest in the event, so we can calculate the contribution of users to the platform (platform revenue) with relevant parameters. For both learning models, we evaluate their performance by analyzing their regrets, which measure the deviation in the resulting utility of this model from the theoretically optimal utility of the original model in which all parameters are known in advance.

Specifically, the contributions of this paper are summarized as follows.

  • We formulate the sequential dynamic event recommendation with feedback (SDERF) problem, which can not only obtain users’ feedback but also continuously recommend a set of sequential events for users.

  • For the SDERF problem, we develop two variants of this problem, namely, an online learning model with no contextual information, in which all users share similar intrinsic interests, and an online learning model with contextual information, in which different users have different levels of tolerance for unsatisfactory recommendations.

  • For the online learning model, we design an exploration–exploitation algorithm based on UCB and show that the regret is bounded by O(T+mTlogT+m2T-3), where m=|V| is the total number of events and T is the total number of rounds as well as the total number of users.

  • For the online learning model with contextual information, we design an algorithm based on UCB and show the regret bounded by Olog(mT)+m, where m=|V| is the total number of events and T is the total number of rounds as well as the total number of users.

  • We conduct extensive experiments on synthetic and real datasets to show the effectiveness and efficiency of our proposed algorithms.

The remainder of the paper is organized as follows. We provide an overview of the related work in Section 2. The model statement is introduced in Section 3. The learning models with/without contextual information are proposed in 4 Online learning model, 5 Online learning model with contextual information, respectively. In Section 6, we evaluate the performance of our algorithms through experiments. Finally, we conclude the paper in Section 7.

Section snippets

Related work

In this section, we summarize related work from two categories: EBSNs and the multi-armed bandit problem.

Model statement

In this section, we borrow the setting and notation of [42] and present a formal statement of our basic model. Table 1 lists the notations used in this paper.

Assume there are m different events V={i|1im,iN} in the EBSN platform, waiting for recommendations to users. We assume users log on to the platform sequentially at time t=1,,T, and we use the same symbol to denote the user for simplicity, which means user t arrives at time t. For each user t, the platform determines a sequence of

Model statement

In the previous section, we find that the best recommendation sequence can be determined by simply sorting scores of all events based on Theorem 1. However, in practice, the parameters in the basic model are all unknown; thus, it is necessary to discuss how to obtain them through users’ feedback and provide the optimal recommendation sequence.

In this section, we consider the online version of the problem, where users arrive sequentially, receive the events recommended by the platform, and give

Model statement

In the previous section, we discuss the problem where parameters of w and q are unknown but fixed for all users. This may not be the case in reality, where different users have different levels of tolerance for unsatisfied recommendations, which means w and q will be different in different rounds. In this section, considering the contextual information of users, we suppose that w and q depend on specified features of users, and u similarly depends on features of events, and we consequently try

Experimental evaluation

In this section, we first present evaluation criteria, datasets and the experimental environment; then, we analyze the experimental result on synthetic and real datasets. Finally, we summarize the proposed algorithms.

Conclusion

In this work, we investigate the problem of sequential dynamic event recommendation with feedback (SDERF). We first model the SDERF problem and develop a basic algorithm with known parameters, with the optimal recommendation sequence. Then, we propose two variants of this problem, namely, an online learning model with no contextual information, in which all users share similar intrinsic interests, and an online learning model with contextual information, in which different users have different

CRediT authorship contribution statement

Yuan Liang: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Supervision, Validation, Visualization, Writing - original draft,Writing - review & editing. Chunlin Huang: Formal analysis, Software, Validation, Visualization,Writing - review & editing. Xiuguo Bao: Writing - review & editing. Ke Xu: Funding acquisition,Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

We are grateful to the anonymous reviewers for their constructive comments. The work of Yuan Liang, Chunlin Huang, Xiuguo Bao and Ke Xu is partially supported by the National Science Foundation of China (NSFC) under Grant Nos. 71531001 and 61421003; and the Foundation for Young Scientists of National Computer Network Emergency Response Technical Team/Coordination Center of China under Grant No. 2019Q23. Yuan Liang is the corresponding author of this paper.

References (45)

  • Y. Mo et al.

    Event recommendation in social networks based on reverse random walk and participant scale control

    Future Generation Computer Systems

    (2018)
  • X. Liu, Q. He, Y. Tian, W.-C. Lee, J. McPherson, J. Han, Event-based social networks: Linking the online and offline...
  • J. She, Y. Tong, L. Chen, Utility-aware social event-participant planning, in: Proceedings of the 2015 ACM SIGMOD...
  • A.M. Ahmed et al.

    Event-based mobile social networks: Services, technologies, and applications

    IEEE Access

    (2014)
  • Y. Liang

    Complex dynamic event participant in an event-based social network: A three-dimensional matching

    IEEE Access

    (2019)
  • Z. Qiao, P. Zhang, C. Zhou, Y. Cao, L. Guo, Y. Zhang, Event recommendation in event-based social networks, in:...
  • Y. Liang et al.

    Friendship-aware task planning in mobile crowdsourcing, Frontiers of Information Technology & Electronic

    Engineering

    (2017)
  • J. Cao et al.

    Multi-feature based event recommendation in event-based social network

    International Journal of Computational Intelligence Systems

    (2018)
  • J. She et al.

    Conflict-aware event-participant arrangement and its variant for online setting

    IEEE Transactions on Knowledge & Data Engineering

    (2016)
  • H. Yin et al.

    Joint event-partner recommendation in event-based social networks

  • Z. Wang et al.

    Toward the new item problem: context-enhanced event recommendation in event-based social networks

  • X. Wang et al.

    Unifying virtual and physical worlds: Learning toward local and global consistency

    ACM Transactions on Information Systems (TOIS)

    (2017)
  • C. Chen et al.

    Conflict-aware weighted bipartite b-matching and its application to e-commerce

    IEEE Transactions on Knowledge & Data Engineering

    (2016)
  • Y. Tong et al.

    Bottleneck-aware arrangement over event-based social networks: the max-min approach

    World Wide Web Journal

    (2016)
  • N. Armenatzoglou et al.

    Real-time multi-criteria social graph partitioning: A game theoretic approach

  • W. Zhang, J. Wang, A collective bayesian poisson factorization model for cold-start local event recommendation, in:...
  • L. Nie et al.

    Learning from multiple social networks

    Synthesis Lectures on Information Concepts, Retrieval, and Services

    (2016)
  • K. Li et al.

    On social event organization

  • Y.-C. Sun et al.

    A novel social event recommendation method based on social and collaborative friendships

  • B. Li et al.

    An event recommendation model using elm in event-based social network

  • I. Boutsis, S. Karanikolaou, V. Kalogeraki, Personalized event recommendations using social networks, in: Proceedings...
  • X. Ji, Z. Qiao, M. Xu, P. Zhang, C. Zhou, L. Guo, Online event recommendation for event-based social networks, in:...
  • Cited by (25)

    • Optimization-assisted personalized event recommendation for event-based social networks

      2023, Advances in Engineering Software
      Citation Excerpt :

      Finally, they tested the effectiveness of their suggested algorithms on actual and artificial datasets. In 2022 Liang et al [37] have created an activity similarity matrix and activity similarity graph, and defined a weighted coverage on the maximum similarity graph, after computing the activity similarity based on context. They also suggested a new index to assess the fairness of a suggestion list, as well as a greedy algorithm that approximates a ratio.

    • Temporal Density-aware Sequential Recommendation Networks with Contrastive Learning

      2023, Expert Systems with Applications
      Citation Excerpt :

      For example, Sun et al. (2021) integrate context-aware seq2seq translation architecture into SR. Liang et al. (2021) propose a sequential dynamic event recommendation method based on social networks. Gan and Ma (2022) propose DeepInteract to learn the interaction of multi-view features of both item profiles and user behaviors for SR.

    • Improved artificial bee colony algorithm-based path planning of unmanned autonomous helicopter using multi-strategy evolutionary learning

      2022, Aerospace Science and Technology
      Citation Excerpt :

      Therefore, it is particularly important to estimate the expected value of this evolutionary strategy. The upper confidence bound (UCB) algorithm is one of reinforcement learning algorithms based on index value, which consists of two stages: exploration and exploitation [58]. By using existing knowledge and exploring the external environment, the UCB algorithm can make optimal decisions quickly.

    View all citing articles on Scopus
    1

    https://hbr.org/2014/10/the-value-of-keeping-the right-customers.

    View full text