Quote from: OgNasty on March 19, 2018, 12:44:57 AM

Would be interesting to see how this correlates with price over a longer length of time. You only have 1 day of info? Seems like you'd be able to go back and get historical data for this.

No we collect and analyse data since 24th february, it collects data each minutes 24/7. For historical data we already downloaded some data from different sources but it's not in use right now. It's part of the optimisation once we adapt our algorithm to work with technical data (volume, average, etc.)

Quote from: samp123 on March 19, 2018, 08:44:02 AM

I tried your algorithm at https://bitmood.net/try, the sentiment worked well but your entity tagging seems a little off. When enabling Entity-Sentiment Analysis it didn't pick up the entity "Bitcoin', for example.

I've used Stanford Core NLP in the past for entity tagging, it's a pre-trained model built by Stanford research teams. Perhaps you'll find it useful as well: https://stanfordnlp.github.io/CoreNLP/

Thanks for your feedback and the reference. Entity tagging needs some optimisation and it's not what we use for the sentiment at the moment, we use the overall sentiment of each tweets. However we integrated the entity features in the API for a test phase.
Our algorithm already use Stanford CoreNLP to determine entities and coreferences associated with each entity. It then determines the sentiment on these (at character level). The problem is that it is a bit simple for now - it only looks for capitalised nouns which are also subjects.

I'll soon update this so it also picks up lowercase sentence objects (not subjects) like bitcoin as an entity. Then the process remains the same - coreference -> new active sentence generation -> sentiment analysis.

Will post when this is updated.