Source-linked AI summary
Popularity Prediction in Microblogging Network: A Case Study on Sina Weibo
Peng Bao, Hua-Wei Shen, Junming Huang, Xueqi Cheng
TL;DR
Popularity prediction is difficult and existing methods largely overlook the structure of early adopters’ propagation networks. Using Sina Weibo, the paper measures early retweet-path structure and incorporates it with early popularity, finding that structural diversity significantly improves prediction accuracy.
Problem
Popularity prediction matters for social-media users and hosts, but existing methods pay little attention to structural characteristics of early adopters’ propagation networks.
Method
The paper measures early retweet-path link density and diffusion depth, then combines each with early popularity to predict final tweet popularity.
Results
The proposed approaches significantly reduce popularity-prediction error, with diffusion depth performing better than link density and the early-popularity baseline.
Takeaways & Limitations
Structural diversity among early adopters provides evidence about final popularity and offers a perspective for building more accurate prediction models.
Abstract
from arXiv · showhide
Predicting the popularity of content is important for both the host and users of social media sites. The challenge of this problem comes from the inequality of the popularity of con- tent. Existing methods for popularity prediction are mainly based on the quality of content, the interface of social media site to highlight contents, and the collective behavior of user- s. However, little attention is paid to the structural charac- teristics of the networks spanned by early adopters, i.e., the users who view or forward the content in the early stage of content dissemination. In this paper, taking the Sina Weibo as a case, we empirically study whether structural character- istics can provide clues for the popularity of short messages. We find that the popularity of content is well reflected by the structural diversity of the early adopters. Experimental results demonstrate that the prediction accuracy is signif- icantly improved by incorporating the factor of structural diversity into existing methods.
1. INTRODUCTION
Popularity prediction can help users filter online content and help advertisers optimize revenue, but it is difficult because popularity is highly asymmetric and affected by many factors. Existing work has studied diffusion, early popularity, user behavior, and classification, while giving little attention to propagation-network structure.
- Popularity prediction supports content filtering for users and differential pricing or placement decisions for advertisers.
- The task is challenging because online-content popularity is broadly distributed and influenced by numerous factors.
- Prior approaches extrapolated early popularity, modeled voting through interestingness and visibility, or formulated prediction as classification.
- Existing methods paid little attention to structural characteristics of content-propagation paths.
2. PROBLEM STATEMENT
The paper predicts a tweet’s popularity at a reference time from forwarding information observed before an earlier indicating time. Popularity is measured by the tweet’s retweet count at the relevant time.
- The task predicts p(t_r), a tweet’s popularity at reference time t_r, using forwarding information available before indicating time t_i.
- Popularity p(t) is measured as the number of times a tweet is retweeted at time t.
3. FINDINGS AND METHODS
The paper analyzes early retweet-path structure through link density and diffusion depth, then incorporates these characteristics with early popularity in prediction models. It finds that lower link density and greater diffusion depth are associated with higher final popularity.
- FINDINGS AND METHODS: Early retweet paths use link density and diffusion depth to characterize the structural diversity of adopters.Link density is the ratio of followship links to possible links; diffusion depth is the longest path from submitter to an early retweeter.
- FINDINGS AND METHODS: Final popularity has a strong negative linear relationship with link density and a strong positive near-linear relationship with diffusion depth.
- FINDINGS AND METHODS: The proposed models combine early popularity with link density or diffusion depth to estimate logarithmic final popularity.
- FINDINGS AND METHODS: The baseline predicts final popularity using early popularity alone.
4. EXPERIMENTS
Experiments use a large Sina Weibo dataset and evaluate predictions with RMSE and MAE. Incorporating link density reduces error relative to the early-popularity baseline, while incorporating diffusion depth performs even better.
- EXPERIMENTS: The dataset contains 16.6 million tweets, with 75% used for training and 25% for testing.
- EXPERIMENTS: The diffusion-depth approach performs better than the link-density approach, which significantly reduces prediction error versus the baseline.Predictions are evaluated with RMSE and MAE; Table 1 reports the comparison.
- EXPERIMENTS: The learned coefficients for link density and diffusion depth are α2 = 0.04 and β2 = 0.07, respectively.
- EXPERIMENTS: The results indicate that low link density and long diffusion paths correspond to wider spreading across the network.
5. CONCLUSIONS
Structural characteristics of early diffusion paths provide strong evidence for final popularity and improve prediction of short-message popularity on Sina Weibo.
- Experiments show that incorporating early adopters’ link density and diffusion depth significantly reduces popularity-prediction error.Low link density and deep diffusion usually indicate wider spreading, reflecting the greater reach of diverse early adopters.