TY - GEN
T1 - What's in a Caption?
T2 - 3rd International Conference on Vocational Education and Electrical Engineering, ICVEE 2020
AU - Hidayati, Shintami Chusnul
AU - Prayogo, Raden Bimo Rizki
AU - Karuniawan, Satria Ade Veda
AU - Hasan, Mhd Fadly
AU - Anistyasari, Yeni
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/10/3
Y1 - 2020/10/3
N2 - In the past few years, social media has become an integral part of modern society. It also hassurfaced as an influential tool that helps a business or individual in gaining identity and reputation. Predicting the popularity of images before they are posted on social media thus may have a profound impact to reveal individual preference and public attention. However, an accurate prediction is a challenging task, mainly on account of factors that play a part in this. Previous studies, although achieve favourable results, overlook one unique characteristic of semantics in textual metadata, i.e., the language modeling, to better model the context information of a post. To that end, wepropose to exploit the language modeling features together with user profile and post metadata features. The language model features are extracted by utilizing the probability of word occurrence, while the user profile and post metadata features are provided as attributes by the original data source. Several state-of-the-art statistical modeling techniques are employed to investigate the performance of the proposed features on different estimation procedures. Experiments on a large-scale Flickr dataset demonstrate the benefits of the proposed features on predicting the popularity of social media posts.
AB - In the past few years, social media has become an integral part of modern society. It also hassurfaced as an influential tool that helps a business or individual in gaining identity and reputation. Predicting the popularity of images before they are posted on social media thus may have a profound impact to reveal individual preference and public attention. However, an accurate prediction is a challenging task, mainly on account of factors that play a part in this. Previous studies, although achieve favourable results, overlook one unique characteristic of semantics in textual metadata, i.e., the language modeling, to better model the context information of a post. To that end, wepropose to exploit the language modeling features together with user profile and post metadata features. The language model features are extracted by utilizing the probability of word occurrence, while the user profile and post metadata features are provided as attributes by the original data source. Several state-of-the-art statistical modeling techniques are employed to investigate the performance of the proposed features on different estimation procedures. Experiments on a large-scale Flickr dataset demonstrate the benefits of the proposed features on predicting the popularity of social media posts.
KW - affective computing
KW - popularity prediction
KW - social media
KW - textual pattern
UR - http://www.scopus.com/inward/record.url?scp=85096641136&partnerID=8YFLogxK
U2 - 10.1109/ICVEE50212.2020.9243175
DO - 10.1109/ICVEE50212.2020.9243175
M3 - Conference contribution
AN - SCOPUS:85096641136
T3 - Proceeding - 2020 3rd International Conference on Vocational Education and Electrical Engineering: Strengthening the framework of Society 5.0 through Innovations in Education, Electrical, Engineering and Informatics Engineering, ICVEE 2020
BT - Proceeding - 2020 3rd International Conference on Vocational Education and Electrical Engineering
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 3 October 2020 through 4 October 2020
ER -