Social Image Sentiment Analysis by Exploiting Multimodal Content and Heterogeneous Relations

Main Article Content

Mohd Amer, Mohammed Jameel Hashmi, C. Kishor Kumar Reddy

Abstract

Sentiment analysis, because to its potential in comprehending people's attitudes and emotions in the context of social big data, is gaining a lot of interest. As massive data emerges on social platforms with numerous manifestations, traditional sentiment analysis approaches that concentrate on a single modality become inadequate. It is argued in this article that multimodal learning techniques, which focus only on the channels rather than the regions, are necessary to capture the links between images and texts. Furthermore, social photos on the social platforms are interconnected in a variety of linkages that are likewise conducive to sentiment classification but are mostly ignored by current research. To enhance the effectiveness of multimodal sentiment analysis, we suggest a heterogeneous relational model based on the user's attention. To learn the combined image-text representation from the viewpoint of content information, we propose a progressive dual attention module to first capture the connections between picture and text. In this work, we suggest using a channel attention schema to draw attention to semantically rich picture channels, and we develop a region attention schema to draw focus to emotional regions based on the attended channels. We next build a heterogeneous relation network and augment a graph convolutional network to integrate content information from social settings as supplements to develop high-quality representations of social pictures. Experiment findings show that our suggested model is superior to the state of the art, and our approach is fully examined on two benchmark datasets.

Article Details

Section
Articles