Unlike films and songs, whose subjects are not always direct representations of their content, academic journal articles come with straightforward titles that communicate research objectives, results, and / or methodologies. Therefore, title generation for an academic journal article can be viewed as a summarization task based on the body of the article. With this in mind, my teammate and I compared performances of a statistical model (TF-IDF), and a deep neural network (RNN) approach to summarization, using a NeurIPS dataset.
My main contribution includes training our own GloVe word embedding, setting up and training the RNN model with an attention mechanism. Check out our presentation slides below, and our code here.
Karina Huang 