Sentiment data starts out as plain social media posts aggregated across all the major social media platforms. It is then run through one or many machine learning models, typically NLP models, that try to understand what is being said in the social media post. You can then aggregate this data at a topic level, or a ticker level to start to gain interesting insight in real time.
Trading with sentiment data is the process of trading equities (stocks) or crypto based on what people are saying about a particular stock or token.
For example, let's say there's a new crypto token launched and there is some buzz around it on Reddit where the first ever post about the token was published. Using a tool to pick up on new stocks could read this post, know that it was about a new stock, understand whether the article was positive or negative as well as read through all the comments about it. In aggregate, you'd be able to tell how something was being talked about as well as the relative reach of any particular post.
Adding other metrics like how many shares or likes a post have can also tell you the amount of attention a particular post receives. When you combine the share data along with the sentiment data, you can then deduce some kind of momentum relative to a baseline. In our example above, there is no baseline, but if you apply this to very well known stocks like $TSLA or $BTC then you can start to pick up on deviations from normal. So for instance if there is some big news about Bitcoin (let's say China banning mining) then there would be a lot of negative chatter on social networks about this. If you use a product that is constantly monitoring social networks for events like this, you could potentially pick up on it right before everyone else does and make better trades by incorporating sentiment data.
Sentiment data is part of a larger umbrella term called alternative data or alt data for short. You can mix and match different data sets and (with back testing of course) try to come up with something that beats baseline performance.