ObjectiveBitcoin - A bitcoin sentiment indicator

[Gwsmyda]

Hello everyone,

I released my website www.objectivebitcoin.com earlier today, and I thought it might be of particular use to speculating efforts.

Basically what my service does is collect submitted content from various sites on the interwebs, classifies each bit of information as positive or negative, and compiles its findings into an indicator value.  The indicator is a z-score of averaged sentiment from the past x hours when compared to the distribution of the past y days.  X and Y are variable, as I'm still working on fine tuning the indicator to better correlate with price changes.

Right now the indicator is updating every hour, but I have plans to increase that resolution, as well as develop an API.

Let me know what you think - I'd be happy to hear any feedback you have.

[NothinG]

Interesting! PM me if there are any resources I could toss at it to collect more data.
Would be cool to see the list of information that it's pulling and what it's considering as positive/negative.

[Gwsmyda]

Would be cool to see the list of information that it's pulling and what it's considering as positive/negative.

That's definitely something I want to implement.  When I first started out, I was reading every tweet that was coming through, and you get some real gems.

[NothinG]

Would be cool to see the list of information that it's pulling and what it's considering as positive/negative.

That's definitely something I want to implement.  When I first started out, I was reading every tweet that was coming through, and you get some real gems.

Tell me about it. I was being obnoxious around $21+, and started going crazy trying to get my friends to just LOOK at it $100+. So, you can kinda imagine the resent comments. They say "Bitcoin is crashing" when in reality Bitcoin has never been better.

[Chalkbot]

Do you also apply weight to each piece of submitted content in accordance with how many people might see that thing, or some other way?

[BubbleBoy]

A past graph would be nice, to put things in context. Maybe add some distinguished points with important news stories, like google does in trends.

[Gwsmyda]

Do you also apply weight to each piece of submitted content in accordance with how many people might see that thing, or some other way?

Yes, the functionality to do that is in place, but I'm still revising things and will probably update that pretty soon.  I was thinking about calculating a "global reach" or something like that.

[Gwsmyda]

A past graph would be nice, to put things in context. Maybe add some distinguished points with important news stories, like google does in trends.

That's definitely coming up!  I was not aware that google trends did that, so I'll check that out for inspiration.

[NikolaTesla]

All you truly need is this: http://www.whatdoestheinternetthink.net/bitcoin
 

[Gwsmyda]

All you truly need is this: http://www.whatdoestheinternetthink.net/bitcoin
 

Holy crap - how did I miss that!?

That positive-to-negative ratio is about the average that I'm getting, which is comforting.  I'm hoping that the fluctuations in my results will be significant.

Either way though, the search for bitcoin returns ~300 hits, while I'm classifying a few times that amount per hour.  That number will increase as I add more places for the program to search.  My sentiment classifiers are also trained on the specific type of information that they'll be reading, e.g. tweets.

[im3w1l]

Very, very cool.

[Aristotle]

Interesting. Did you train the classifier? How did you get the labels (positive/negative)? Are you using k-grams, bag of words, or something else as features?

[Gwsmyda]

Interesting. Did you train the classifier? How did you get the labels (positive/negative)? Are you using k-grams, bag of words, or something else as features?

I manually classified a few thousand items and trained the classifier(s) on those.  As of now I'm using bag of words, bigrams, and trigrams as features - classification accuracy seems to be pretty decent.

[Gwsmyda]

Now with 100% more charts!