Efficient Visual Sentiment Prediction Approaches Using Deep Learning Models

• Published in: Knowledge Graphs and Semantic Web Vol. 1459; pp. 260 - 272
• Main Authors: Chandrasekaran, Ganesh, Hemanth, D. Jude
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2021; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Image sentiment prediction; Natural Language Processing (NLP); Transfer learning and deep learning
• ISBN: 9783030913045; 303091304X
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-030-91305-2_20

Sentiment analysis of images is becoming very important with the increase in the use of social media by people. Many existing works in the literature focus on textual sentiment analysis with the text being extracted from the social media sites like Twitter, Facebook, Amazon and Movie reviews, etc. But sharing of images and videos through social media is increasing compared to text. Images reflect the sentiment in a much better way compared to the text and thus are preferred in analyzing the sentiment. So, there is a need to develop a robust model to carry out image sentiment prediction. In this paper, we employ a transfer learning model based on VGG-16 architecture to carry out the image sentiment analysis. The dataset employed is the Crowdflower database that has more than 15,000 images (Twitter) URLs with its polarity label (Positive, Negative). The proposed model can handle a large set of data as it is based on deep learning and there is no need for explicit feature extraction. The results show the effectiveness of our proposed model in performing sentiment prediction on images.