Meme Classification and Offensive Content Detection Using Multimodal Approach
Tvisha Modi, Esha Shah, Sachi Shah, Janki Kanakia, Manisha Tiwari · 2024
Hate speech and offensive content on social media platforms, often conveyed through memes, pose significant challenges for content moderation and community well-being. This paper classifies hateful memes and proposes a novel approach leveraging a multimodal approach. The model uses Long Short-Term Memory (LSTM) networks to handle textual material and VGG16 convolutional neural networks (CNNs) to analyse image data. Additionally, it enhances platform safety by enabling the detection of harmful content in memes that may evade traditional detection methods by obscuring or altering text. Moreover, the suggested methodology facilitates the obscurement of objectionable words, inside the meme, allowing them to be shared on multiple social media platforms without promoting hate speech or offensive discourse. This approach aims to create a balance between content moderation and free expression on social media platforms, resulting in a better, healthier and safer online environment.