Our project focuses on creating an accurate text summarizer for news articles. We developed an API that uses various NLP models to acquire a summary of an entire news article. While the API can be potentially used for various other cases including, Movie and book summaries, e-commerce product review summaries, and so on. Our product Gist focuses on a small part of it for newspaper summarization. This is why we have open-sourced our project for it to be used by people as per their requirements and scope. The main difference between our summarize and other summarizes already in the market is that ours is an abstractive type rather than extractive, which means, it focuses on creating and framing it's own summaries rather than just focusing on points which are relevant and copy pasting them in the summary.
Taking part in a 36-hour-long hackathon is no easy task. However, our main hurdle came into the picture when we were trying to figure out an optimal model that we could use to summarize our articles. Finding a model that works both efficiently and summarizes our articles abstractively was quite an issue. However, Hugging face transformers came for the clutch. It helped us use it's pre-trained models thus saving a lot of our time in training and retraining them to fit our product. It also resolved several of our queries looming around the technology stack that we were using thanks to it's amazing documentation.
Discussion