Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A layer-wise deep stacking model for social image popularity prediction

A layer-wise deep stacking model for social image popularity prediction In this paper, we present a Layer-wise Deep Stacking (LDS) model to predict the popularity of Flickr-like social posts. LDS stacks multiple regression models in multiple layers, which enables the different models to complement and reinforce each other. To avoid overfitting, a dropout module is introduced to randomly activate the data being fed into the regression models in each layer. In particular, a detector is devised to determine the depth of LDS automatically by monitoring the performance of the features achieved by the LDS layers. Extensive experiments conducted on a public dataset consisting of 432K Flickr image posts manifest the effectiveness and significance of the LDS model and its components. LDS achieves competitive performance on multiple metrics: Spearman’s Rho: 83.50%, MAE: 1.038, and MSE: 2.011, outperforming state-of-the-art approaches for social image popularity prediction. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png World Wide Web Springer Journals

A layer-wise deep stacking model for social image popularity prediction

Download PDF
17 pages

Loading next page...
 
/lp/springer_journal/a-layer-wise-deep-stacking-model-for-social-image-popularity-aUyQ3oKR36
Publisher
Springer Journals
Copyright
Copyright © 2018 by Springer Science+Business Media, LLC, part of Springer Nature
Subject
Computer Science; Information Systems Applications (incl.Internet); Database Management; Operating Systems
ISSN
1386-145X
eISSN
1573-1413
DOI
10.1007/s11280-018-0590-1
Publisher site
See Article on Publisher Site

Abstract

In this paper, we present a Layer-wise Deep Stacking (LDS) model to predict the popularity of Flickr-like social posts. LDS stacks multiple regression models in multiple layers, which enables the different models to complement and reinforce each other. To avoid overfitting, a dropout module is introduced to randomly activate the data being fed into the regression models in each layer. In particular, a detector is devised to determine the depth of LDS automatically by monitoring the performance of the features achieved by the LDS layers. Extensive experiments conducted on a public dataset consisting of 432K Flickr image posts manifest the effectiveness and significance of the LDS model and its components. LDS achieves competitive performance on multiple metrics: Spearman’s Rho: 83.50%, MAE: 1.038, and MSE: 2.011, outperforming state-of-the-art approaches for social image popularity prediction.

Journal

World Wide Web – Springer Journals

Published: May 28, 2018

References

  • On the prediction of flickr image popularity by analyzing heterogeneous social sensory data
    Aloufi, S; Zhu, S; El, SA
  • Internet recommendation systems
    Ansari, A; Essegaier, S; Kohli, R
  • Random forests
    Breiman, L
  • A cross-media public sentiment analysis system for microblog
    Cao, D; Ji, R; Lin, D; Li, S
  • Support-vector networks[J]
    Cortes, C; Vapnik, V
  • Extremely randomized trees
    Geurts, P; Ernst, D; Wehenkel, L
  • Flickr group recommendation using rich social media information
    Guo, C; Li, B; Tian, X
  • The bayesian lasso
    Park, T; Casella, G
  • Towards cross-domain learning for social video popularity prediction
    Roy, SD; Mei, T; Zeng, W; L, S
  • Statistical methods
    Snedecor, GW; Cocheran, WG
  • Analyzing Flickr metadata to extract location-based information and semantically organize its photo content
    Spyrou, E; Mylonas, P
  • Dropout: a simple way to prevent neural networks from overfitting
    Srivastava, N; Hinton, G; Krizhevsky, A
  • Sparse multi-graph embedding for multimodal feature representation
    Wang, s; Guo, W
  • Dual graph regularized NMF model for social event detection from flickr data
    Yang, Z; Li, Q; Lu, Z; Gong, Z; Liu, W