On Organizing Online Soirees with Live Multi-Streaming

Chih-Ya Shen, Cedric Parfait Kankeu Fotsing, De-Nian Yang, Yi-Shin Chen, Wang-Chien LeeAAAI 2018 · DOI

Introduction

Live streaming platforms have evolved from simple broadcast tools into rich social environments. Modern viewers often engage in "Live Interactions" and "Multi-Streaming"—watching multiple perspectives of a single event simultaneously. However, organizing a shared viewing experience, or an "online soiree," remains a manual and complex task. Current recommendation systems typically focus on individual preferences, ignoring the social dynamics and content diversity required for a successful group event.

To address this, we formalize the Social-aware Diverse and Preferred Live Streaming Channel Query (SDSQ). This framework aims to select a socially tight group of friends and a diverse set of live channels that satisfy the collective interests of the group.

Conceptual model of the SDSQ framework showing the interaction between the Channel Layer and Social Layer.

The SDSQ Framework and SDSSel Algorithm

The SDSQ problem is computationally challenging, proven to be NP-hard and inapproximable in its standard form. The goal is to maximize the average dissimilarity (diversity) of selected channels while ensuring social tightness (users within $h$ hops) and meeting minimum interest thresholds for each viewer.

We propose SDSSel, an efficient 2-approximation algorithm utilizing a "relax-and-tailor" strategy:

  1. Relaxation: The algorithm identifies "2h-dense" sets, which are slightly looser social circles, making the search space manageable for large networks.
  2. Selection: It iteratively prunes channels with low "incident diversity" while maintaining group interest constraints.
  3. Tailoring: A post-processing step refines the group to meet strict social tightness requirements while preserving the approximation ratio for diversity.

Experimental Evaluation

The effectiveness of SDSSel was evaluated using diverse datasets, including a custom Twitch user study (432 users), Yelp (1.5M users), Douban (5M users), and a combined Twitch/Twitter dataset (6M viewers).

Key Results

Our evaluation demonstrates that SDSSel significantly outperforms existing baselines and manual selection across several metrics:

  • User Satisfaction: In user studies, SDSSel achieved over 50% user satisfaction, far exceeding baselines like TrustSVD and GD-LM-MIN.
  • Feasibility: Unlike manual selection or standard recommendation models, SDSSel maintains a 100% feasibility ratio, ensuring all social and interest constraints are met.
  • Efficiency: SDSSel processes millions of users in under 90 seconds, whereas baseline methods often fail to scale beyond small groups.
  • Diversity: By maximizing average dissimilarity, the algorithm ensures a high-quality, non-redundant selection of content.
Performance metrics of SDSSel: (a) critical factors for users, (b) user satisfaction vs. baselines, (c) average diversity, and (d) feasibility ratio.

Conclusion

The SDSQ framework and the SDSSel algorithm provide a robust solution for automating the organization of online soirees. By balancing social tightness, user preference, and content diversity, SDSSel enables a more engaging and less "lonely" live streaming experience. Future research will explore dynamic queries to accommodate real-time changes in user availability and stream status.

Summary updated

BibTeX
@inproceedings{shen2018organizing,
  author = {Chih-Ya Shen and Cedric Parfait Kankeu Fotsing and De-Nian Yang and Yi-Shin Chen and Wang-Chien Lee},
  title = {{On Organizing Online Soirees with Live Multi-Streaming}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1609/aaai.v32i1.11310},
}
Figure