My model for detecting if a meme is hateful/dank
We all come across memes over the internet. They are the craze of today! But some of them are not really for everyone. They are deliberately aimed to hurt religous sentiments which people tend to take offense to. Here's an example:
Not very tasteful I know. But then there are also harmless memes like this:
Classifying them automatically poses a huge challenge and can be framed as a Deep Learning problem.
Our solution uses a CNN which extracts features from the image, BERT, which is used to get embeddings from the meme caption. The concatenated output of the two is then passed through as dense layer to finally predict the output
You can run the notebook in the above repo using colab.