pietrocagnasso/paper-title-generation

Title generation task

This demo shows in action the models we developed for the exam Deep Natural Language Processing during a.y. 2022-2023.

The demo gets as inputs the concatenation of the highlight of a paper and the abstract of the same paper to produce a suitable title.

The model we used for this task is BART, for the implementation details refer to this repo on GitHub.

The input used in this Gradio app is different from the one used in the original implementation. We require to add at the beginning of the sentences the topic they refer to (CS, AI, BIO), in this way we can try to use the best model available. If not topic is received the defulat model is the one trained on the mix of the three topics. Refer to examples to understand how to write the input.