Abstract | ||
---|---|---|
This paper addresses the problem of predicting future image popularity on social networks by considering the changes of image popularity over time. For this problem, we collect information about an image within an hour of upload and keep track of its popularity for one month to predict its future popularity (e.g., after a day, a week, a month). For the prediction of popularity, we employ three features: social context (i.e., user information), the image's semantics and the image's early popularity. We propose a novel approach to extract the semantic of images, based on well established natural language processing and clustering techniques. Using a Gaussian Naive Bayes classifier, we predict the future popularity of images using such social context, image semantics, and early popularity. The results show that the accuracy of the classifier reaches 90% on average in predicting future popularity; image semantics is the only feature that increases popularity predictions accuracy along the timeline. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1145/2955129.2955154 | MISNC |
DocType | Citations | PageRank |
Conference | 2 | 0.38 |
References | Authors | |
10 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Khaled Almgren | 1 | 2 | 0.38 |
Jeongkyu Lee | 2 | 285 | 24.82 |
Minkyu Kim | 3 | 22 | 9.55 |