This small webapp measures Twitter sentiment along seven psychological axes based on a paper by Bollen, 2010, titled Twitter mood predicts the stock market.
It's a work in progress and a MVP.
As a webapp it is similar to A World of Tweets by frogdesign, tweetping.net by Franck Ernewein and the hedonometer.org.
Thank you all for this feedback. We were amazed by the quality of it.
Note that we have already modified our newsletter subscription process. It is now opt-in instead of opt-out as it should have been from the start. We apologize for that. We have unsubscribed everyone that signed-in since we posted. If anyone wants to hear from us, please subscribe by going to Settings -> Profile.
We use Superfeedr for feeding the news into the python back-end running on AWS. SQS queues are used to send and receive messages between our web portal and analyzers. The analysis is done in real-time and reported to the web application running on Google App Engine. Amazon Simple Notification Service is used to report to the webapp (and will also be used to send to users subscribed to the API). We use S3 to store the raw documents, but CouchDB for the metadata and the results of our algorithms.
Regarding our algorithms, we use Reuters' OpenCalais as a first step to the NLP, but augment it with our own stuff for the bull and bear extraction.
The webapp uses Knockout.js and all other standard stuff like jQuery and underscore.js. We "build" the javascript using require.js.
What about giving the pros the ability to upload their own documents (and remain private) for BullBear to analyze and report? Also, what about the ability to customize which source of information the system process and which weight each source receive in the bullbear index calculation?
These are all ideas we have on the table for the near future.
If it could do things like that, then maybe selling it as a product that a company can plug their own sources into and let your algos do the crunching and analysis and present the data could be a good move.
We are looking into adding time information. See me reply to jnorthrop. Essentially, we are working on distinguishing between a news stating a fact that occurred in the past, present or future. Also, a time horizon has been requested before.
Not ourselves, but we know people you do. This whole idea was proposed by a commodity trader friend of ours and we decided to give it a shot. This is what came of it. So far, our trader friend likes it.
I agree, predicting purely on lagging information is no good. We had this comment before and that's why we are currently working on distinguishing between a news stating a fact that occurred in the past, present or future. With that, we'll be able to present what the market thinks of the future price movements. For example, an author saying: "we expect copper to surge in the coming months", would be recognized as a future statement and factored in the future bullbear index. Or something like that.
That's smart, and very cool. I recall the dollar signs in my eyes when I thought about the potential in getting a predictive system like this to work -- I hope you guys nail it.
Actually, we are doing clustering, but as you found out it's not perfect. Like you said, we don't want the same news repeated over and over again to be factored in the bullbear index.
I like this idea: Instead of showing 10 most recent articles it's better to show N-most recent articles with S-most recent trend.
Currently the weighting is flat, but that is about to change. What we want to offer, and let me know if that is something of interest to you, is the ability for the user to customize the sources and to customize the weight of each sources. Each traders probably has its own preferred and trusted sources and we want to address that.
It's a work in progress and a MVP.
As a webapp it is similar to A World of Tweets by frogdesign, tweetping.net by Franck Ernewein and the hedonometer.org.