Bitcoin Forum

Hi guys, posting here to get some feedback about a new project me a friend just released in beta version.
It's a cryptocurrency sentiment analysing tool that use deep learning algorithm constantly monitoring tweets and soon news.
Here's the link http://bitmood.net. We already analysing Bitcoin and Iota.
Feel free to give me negative feedback and suggestion, I would love to know if some of you would use it or have ideas about improvement that would make the website more usefull for you.
Thanks  ;)

Whenever you publish a project which uses deep learning,you must explain what is the underlying algorithm used in the process.Otherwise,it becomes difficult for people like me to analyse the data  and figure out how relevant it is.

Also graph could have a small section explaining what those data-points represent.

EDIT : You have this link https://bitmood.net/try which analysis data I suppose? What data does it expects as an input ? You need to put requirements there man!

Thanks for your feedback.

I'll take your advice and we'll explain it on the website soon.

Actually it's a recurrent neural network using LSTM to analyse tweets sentiment. We trained the model with 1 million classified tweets about everything. We plan to get or make a training set based only on trading, our model will be more accurate.

The datapoints represents the number of tweets we analysed, it's categorized by positive, neutral and negative and it's represented on the graph.

The try link just wait for a sentence and return the entity sentiment (if you select entity-based analysis) or the overall sentiment. It's not perfect and we are currently working on algorithm improvement, we expect some feedback in order to know if people would use that kind of information/service. One of our next goal is to cross technical and sentiment analysis in order to be even more accurate about the price prediction per coin.

Sounds great!

So basically you analyse tweets and categorise them as "negative/positive" ? Interesting.What are your data sources ? Random bitcoin accounts or you follow a list of people ?
 
Understood.

Not sure about anybody else but being a programmer myself,I love tinkering with data.(and I love Data Structures/Algorithms in general)

Yes i’s categoized by positive and negative and we calculate the neutral base on the score.

Our datasource is everything that include bitcoin or other coin keyword. Could be a good idea to follow some relevant people. Actually we have a lot of noise in our data and it’s another improvment to clean them.

Are you also a crypto investor? As investor ourself we plan to use our own tool to make decision and get a good overview of each coin.

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.

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/

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.)


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.

i really like to look at sentiments for analysis and review of decision making processes. I am glad you have built a tool. Do you need help with beta testers? let me know I can help in this area

Hello fellow SwissBoo, thanks for your proposal. Yep we need a lot of beta testers  ;D feedback help us keep motivation and improve the tool that people would use and find useful.

The interface of your website is quite easy to look at

I would love to see what would happen if you used this tool and pulled historical data.

You definitely need to include a [cryptocurrency benign analysed] price trend line so that it could be visible how your sentiment line correlates with the base asset's price. 

You're perfectly right and it's in our ToDo list  ;)

Website you have given does not working bro. If there are any separate URL you have created for this beta version? after you have created this thread. I personally admire the person take the project for take crypto currency in sentimental aspects.

Lets focus about the various coins and invest large time on analyze the coins and go with them dude.
Good luck for your work hope we see this website without finding application error bro.

Hi solarion, thanks for your comment. Sorry that you found our website down, we're still optimising our infrastructure. I just upgraded our database to a more powerfull one and we corrected some error. Hope next time you'll find our website up and running bro  ;D