Color-based visual sentiment for social communication

Mayank Amencherla, Lav R. Varshney

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Social media platforms provide rich signal sets to understand the nature of social life, and sentiment analysis techniques have been developed to understand the emotional content of text from sites like Twitter and Facebook. Beyond text however, most social media platforms have images at their core, and communication of images may require quantization. Here, we develop methods and present results on understanding the association between the visual content features of images on the popular social media platform Instagram and the psycholinguistic sentiment of their hashtag descriptors. In particular, we collect several thousand images and analyze several aspects of color to predict image sentiment. These results affirm and clarify several psychological theories on the relationship between color and mood/emotion, such as colorfulness being associated with happiness. The data-driven psychovisual insights into sentiment developed herein can be used to define novel fidelity criteria for designing color quantization schemes.

Original languageEnglish (US)
Title of host publication2017 15th Canadian Workshop on Information Theory, CWIT 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781509060269
StatePublished - Jul 27 2017
Event15th Canadian Workshop on Information Theory, CWIT 2017 - Quebec City, Canada
Duration: Jun 11 2017Jun 14 2017

Publication series

Name2017 15th Canadian Workshop on Information Theory, CWIT 2017


Other15th Canadian Workshop on Information Theory, CWIT 2017
CityQuebec City


  • big data
  • color
  • images
  • sentiment analysis
  • social signals

ASJC Scopus subject areas

  • Software
  • Computer Networks and Communications
  • Information Systems


Dive into the research topics of 'Color-based visual sentiment for social communication'. Together they form a unique fingerprint.

Cite this