Discovering Strategic Behaviors for Collaborative Content-Production in Social Networks

Yuxin Xiao, Adit Krishnan, Hari SundaramWWW 2020 · DOI · OpenReview

Abstract

Some social networks provide explicit mechanisms to allocate social rewards such as reputation based on user activity, while the mechanism is more opaque in other networks. Nonetheless, there are always individuals who obtain greater rewards and reputation than their peers. An intuitive yet important question to ask is whether these successful users employ strategic behaviors to become influential. It might appear that the influencers have gamed the system. However, it remains difficult to conclude the rationality of their actions due to factors like the combinatorial strategy space, inability to determine payoffs, and resource limitations faced by individuals. The challenging nature of this question has drawn attention from both the theory and data mining communities. Therefore, in this paper, we are motivated to investigate if resource-limited individuals discover strategic behaviors associated with high payoffs when producing collaborative or interactive content in social networks. We propose a novel framework of Dynamic Dual Attention Networks or DDAN which models content production strategies of users through a generative process, under the influence of social interactions involved in the process. Extensive experimental results illustrate the effectiveness of our model in modeling user behavior. We make three strong empirical findings. Different strategies give rise to different social payoffs, the best performing individuals exhibit stability in their preference over the discovered strategies, which indicates the emergence of strategic behavior, and the stability of strategy preference is correlated with high payoffs.

Introduction

In any social network, rewards—whether they be citations, likes, or reputation points—are distributed unequally. While some platforms like StackOverflow have explicit point systems, others, such as academic citation networks or social media algorithms, are more opaque. This disparity raises a fundamental question: Do the "winners" of these networks succeed through luck, or have they discovered secret strategies to "game the system"?

This research addresses the challenge of bounded rationality. Humans possess limited time and cognitive resources, yet some navigate massive, complex strategy spaces to reach the top. Previous research has largely focused on what topics people are interested in; this paper shifts the focus to the strategic "how" of content production. We investigate whether resource-limited individuals can discover high-payoff strategies amidst the noise of social interactions.

Methodology: The DDAN Framework

To model these behaviors, we propose the Dynamic Dual Attention Network (DDAN). This generative framework conceptualizes content production as a bipartite graph between authors and the content they create.

The bipartite relationship between authors and content, where content is characterized by a set of strategic attributes.

As shown in the figure above, each piece of content $c_i$ is associated with various attributes that represent strategic dimensions. DDAN treats these strategies as latent probability distributions, assuming that users employ a "mixed strategy"—a blend of pure strategies like "preferential attachment" (popularity) or "topic similarity" (field).

The core of DDAN is its Dual Attention mechanism, which operates across temporal snapshots:

  1. Author-to-Content Attention: This determines the relative contribution of each co-author to the strategy of a specific piece of content.
  2. Content-to-Author Attention: This models how an author’s current strategy is influenced by their own historical behavior and the strategies of their current collaborators.
The DDAN framework modeling the evolution of author and content distributions across time snapshots.

Experimental Setup and Datasets

The model was primarily evaluated using the DBLP academic dataset (1980–2018), focusing on two strategic decisions: Citation strategies (who to cite) and Location strategies (where to publish). Strategy spaces were defined based on four dimensions: Popularity, Field, Familiarity, and Time. The framework was tested on network snapshots, specifically analyzing authors with sustained production histories to observe strategic evolution.

Key Insights and Results

Our analysis yielded several significant findings regarding the nature of success in social networks:

  • The Emergence of Order: Strategy stability is a hallmark of success. The top 10% of performers exhibit high rank correlation in their strategy preferences over time. Conversely, the bottom 90% appear "strategic-less," with inconsistent shifts in their approach.
  • Beating the "Rational" Agent: We compared human performance against an idealized rational agent that always selects the strategy with the highest expected utility. Remarkably, approximately 7% of authors outperformed this rational agent, suggesting they identified high-risk/high-reward niches that a generalized model would overlook.
  • Interdisciplinary Payoffs: High-impact papers often resulted from a strategy of citing highly-cited papers from different fields, confirming that interdisciplinary "bridging" is a high-payoff strategic move.
  • Predictive Superiority: DDAN outperformed several baselines, including Logistic Regression and various Topic Models, in predicting future citations and publications, achieving higher Mean Average Precision (MAP) scores across all time snapshots.

Conclusion

The results demonstrate a clear positive correlation between the stability of a user's strategy and the utility (citations) they receive. The top 1% of authors converge on their winning strategies much faster than their peers, identifying successful heuristics early in their careers. While the current study is focused on academic networks and predefined strategy dimensions, it provides a robust mathematical foundation for understanding how the "elite" in social networks are defined by their strategic consistency.

Summary updated

Abstract

Some social networks provide explicit mechanisms to allocate social rewards such as reputation based on users’ actions, while the mechanism is more opaque in other networks. Nonetheless, there are always individuals who obtain greater rewards and reputation than their peers. An intuitive yet important question to ask is whether these successful users employ strategic behaviors to become influential. It might appear that the influencers ”have gamed the system.” However, it remains difficult to conclude the rationality of their actions due to factors like the combinatorial strategy space, inability to determine payoffs, and resource limitations faced by individuals. The challenging nature of this question has drawn attention from both the theory and data mining communities. Therefore, in this paper, we are motivated to investigate if resource-limited individuals discover strategic behaviors associated with high payoffs when producing collaborative/interactive content in social networks. We propose a novel framework of Dynamic Dual Attention Networks (DDAN) which models individuals’ content production strategies through a generative process, under the influence of social interactions involved in the process. Extensive experimental results illustrate the model’s effectiveness in user behavior modeling. We make three strong empirical findings: (1) Different strategies give rise to different social payoffs; (2) The best performing individuals exhibit stability in their preference over the discovered strategies, which indicates the emergence of strategic behavior; and (3) The stability of a user’s preference is correlated with high payoffs.
BibTeX
@inproceedings{xiao2020discovering,
  author = {Yuxin Xiao and Adit Krishnan and Hari Sundaram},
  title = {{Discovering Strategic Behaviors for Collaborative Content-Production in Social Networks}},
  booktitle = {The Web Conference},
  year = {2020},
  doi = {10.1145/3366423.3380274},
}
Figure