BERT-based Multimodal Aspect-Level Sentiment Analysis for Social Media

Aspect-level sentiment analysis, a sub-task of sentiment analysis, aims to identify the sentiment polarity of a given aspect in a sentence. In recent years, with the diversification of the web, people are no longer satisfied with using text alone to post their status on social media, but also often use images as a way of recording, and the combination of images and text has gradually become an important form of data in the social media domain. However, aspect-based sentiment analysis is currently mostly applied to textual content, which neglects the role of image data in enhancing the robustness of text-based models. This paper proposes a multimodal sentiment analysis framework for social media based on the BERT model. The model is validated by importing text features and image features into the BERT model, which lends itself to extracting the interconnections between text and images. It is shown through experiments that the interaction effects between cross-modal data can be learnt by importing the BERT module, making social media multimodal sentiment classification better, and validating the effectiveness of the model.