Social networks have become a vital part of everybody's life, as users on
popular social networking platforms share millions of images to express their
opinions and personal emotions. Therefore, images have emerged as one of the
most effective methods for transmitting sentiments on social networks. This has
resulted in a solid vision to analyze social network images to predict positive
and negative sentiments from these images. In this paper, an automatic system
based on object recognition is developed by combining InceptionV3 and Long
Short-Term Memory networks for predicting visual sentiments. This system
aims to recognize the salient objects from social network images and predict
their sentiments. Firstly, the InceptionV3 pre-trained CNN network is fine-tuned
to recognize objects from images. After that, the object features are extracted
using the trained network. Finally, a Long Short-Term Memory network is used
to learn sentiments from object features to predict visual sentiment. The
experiment results showed that the proposed system is a more powerful system
for predicting visual sentiments by combining Inception V3 and Long ShortTerm Memory networks. The proposed system achieved 98.2% for predicting
visual sentiments.