Understanding User Topic Preferences across Multiple Social Networks

Ziqing Zhu, Jiuxin Cao, Tao Zhou, Huiyu Min, Bo Liu2021 · arXiv

Abstract

In recent years, social networks have shown diversity in function and applications. People begin to use multiple online social networks simultaneously for different demands. The ability to uncover a user's latent topic and social network preference is critical for community detection, recommendation, and personalized service across social networks. Unfortunately, most current works focus on the single network, necessitating new technology and models to address this issue. This paper proposes a user preference discovery model on multiple social networks. Firstly, the global and local topic concepts are defined, then a latent semantic topic discovery method is used to obtain global and local topic word distributions, along with user topic and social network preferences. After that, the topic distribution characteristics of different social networks are examined, as well as the reasons why users choose one network over another to create a post. Next, a Gibbs sampling algorithm is adopted to obtain the model parameters. In the experiment, we collect data from Twitter, Instagram, and Tumblr websites to build a dataset of multiple social networks. Finally, we compare our research to previous works, and both qualitative and quantitative evaluation results have demonstrated the effectiveness.

Introduction

In the modern digital landscape, users are rarely loyal to just one platform. We might tweet about breaking news on Twitter, share aesthetic snapshots on Instagram, and write long-form hobbyist blogs on Tumblr. While core interests remain consistent, the way users express them shifts to match the "vibe" and functionality of each site.

A recent study by Ziqing Zhu and colleagues from Southeast University introduces a new model designed to bridge these digital silos. By analyzing "overlapping users"—those active on multiple platforms—the researchers developed a way to see the full picture of a user’s interests, distinguishing between their core global preferences and their platform-specific "local" nuances.

The Challenge of Multi-Platform Behavior

Most existing topic modeling research focuses on a single social network. However, users behave differently depending on the platform's affordances. The authors identify two key phenomena:

  1. Consistency: Users often show similar interests across different networks (e.g., a "techie" will post about gadgets everywhere).
  2. Complementarity: Users present different aspects of the same topic on different networks. For example, on one network, a user might discuss a phone's hardware specs, while on another, they focus on the apps and user experience.
Conceptual framework of user behavior across multiple social networks

The challenge lies in creating a model that can fuse this heterogeneous data to understand not just what a user likes, but why they choose a specific network to post about a specific topic.

Methodology: The MSNT Model

The researchers propose the Multiple Social Networks Topic Model (MSNT), a generative framework that discovers a user's latent interests by fusing data from different platforms. Unlike previous models that treat each site in isolation, MSNT identifies "global" topics shared across a user's digital footprint and "local" topics that reflect how those interests are uniquely expressed on specific networks.

Key components of the model include:

  • Global vs. Local Topics: Words in a post can come from a Global Topic distribution (consistent across all sites), a Local Topic distribution (specific to one site), or a Background distribution (common "noise" words).
  • The Switching Variable ($x$): A latent variable acts as a "traffic controller," deciding for each word whether it should be drawn from the global, local, or background pool.
  • Social Network Preference ($\rho_{uz}$): The model explicitly calculates the probability that a specific user will choose a specific social network to discuss a particular topic.

Experimental Evaluation

To test the model, the team built a unique dataset by crawling profiles to find users who linked multiple accounts across Twitter, Instagram, and Tumblr. For the image-heavy Instagram data, the researchers used deep learning-based image annotation to convert images into word tags, allowing visual content to be processed alongside text.

Results and Discussion

The MSNT model was compared against several baselines, including standard LDA and Twitter-specific models like TwitterLDA.

Comparison of Perplexity across different topic models

Quantitative results demonstrated that MSNT achieved significantly lower perplexity and higher likelihood scores than all baselines, indicating it is better at representing the actual data distribution of multi-platform users. Furthermore, the model produced more coherent topics (measured by PMI-score) by successfully separating local nuances from global themes.

Conclusion

The MSNT model provides a robust framework for understanding how users navigate multiple social identities. By accounting for both consistency and complementarity, it offers a more complete profile of user interests than single-network models. Future work will explore the temporal evolution of these preferences and deeper integration of raw multimedia data.

Reference: Zhu, Z., Cao, J., Zhou, T., Min, H., & Liu, B. (2021). Understanding User Topic Preferences across Multiple Social Networks. arXiv:2103.07654v2 [cs.SI].

Summary updated

Abstract

In recent years, social networks have shown diversity in function and applications. People begin to use multiple online social networks simultaneously for different demands. The ability to uncover a user’s latent topic and social network preference is critical for community detection, recommendation, and personalized service across social networks. Unfortunately, most current works focus on the single network, necessitating new technology and models to address this issue. This paper proposes a user preference discovery model on multiple social networks. Firstly, the global and local topic concepts are defined, then a latent semantic topic discovery method is used to obtain global and local topic word distributions, along with user topic and social network preferences. After that, the topic distribution characteristics of different social networks are examined, as well as the reasons why users choose one network over another to create a post. Next, a Gibbs sampling algorithm is adopted to obtain the model parameters. In the experiment, we collect data from Twitter, Instagram, and Tumblr websites to build a dataset of multiple social networks. Finally, we compare our research to previous works, and both qualitative and quantitative evaluation results have demonstrated the effectiveness.
BibTeX
@inproceedings{zhu2021understanding,
  author = {Ziqing Zhu and Jiuxin Cao and Tao Zhou and Huiyu Min and Bo Liu},
  title = {{Understanding User Topic Preferences across Multiple Social Networks}},
  booktitle = {2021},
  year = {2021},
}
Figure