I'd recommend scooting on over to hf.co and checking out some of their open source models. Realistically you can call one, or an ensemble from there in a few lines of code. Not sure what's causing this piece of work - school? work? hobby? But that's a decent place to get something set up quickly. They might even have models fine tuned on stock/economic vocabulary.

If you know how to code with python. Then I would highly recommend trying the spacy module. It’s really awesome and they have a ton of guides and info on their site

Depends on what you're trying to do with this project.
If you already have a news source, implementing pre-trained model from huggingface should be relatively easy.
If you want to fine-tune that model, you will need a dataset of news headlines.
Check out Kaggle, there are some small but publicly available datasets.
I know you can also find data on some obscure github repo, but good luck with that.
If your goal is to implement a fully operational pipeline, you will need not only all the above, but also a way to acquire news in real time. That may mean a scrapper of some news outlets that are of interest. Once again, github is your friend.
That said, don't expect a profit off this alone. Using news data alongside some trading indicators will *maybe* work on paper (ie. backtest) with the right features and optimization, but is unlikely to get live results.
Then again, for a college project that might not be relevant.