Yesterday I discovered a service called OpenAmplify. OpenAmplify is a free API that provides deep, and I mean really deep, textual analysis. Currently they only support English, and I've no idea how long the service will remain free, but the level of detail their analysis provides is pretty darn stunning. As an example of the level of detail it provides, when I used their API to scan this blog, it was able to tell me:
- I'm a male adult (which, ok, to be fair, is pretty easy to guess ;)
- My writing has a high level of decisiveness and is somewhat flamboyant (cool)
- It says I offer guidance "a lot" which is very accurate
- My content is more positive than negative
- My topics include computers, computer programming, html, ajax, etc
- And waaay more info than I can describe here
Along with parsing a URL, you can also check individual strings. So for example, I wrote up a CFC wrapper that allows you to just do this:
<cfset oa = new openamplify("my key")>
<cfset res = oa.parse(text="This is a test of the openamplify system. It is slick!")>
I then dumped out the polarity:
<cfdump var="#res.styles.polarity#">
Polarity values will range from -1 (negative) to 1 (positive). Watch what happens when I change my text to something a bit more excited: "This is a test of the openamplify system. It is super awesome!"
Notice how it went up a bit to reflect the higher level of positivity. I did tests with negative values as well and found the API to be a bit scary in how accurate it was.
Anyway - I'm very impressed by this service so far. Imagine this in a blogging application. You could run a daily report that gathers comments posted in the previous 24 hours and report on the top topics as well as the general mood of the discussions. For an online community, that type of feedback could be killer. If you see a sharp swing to the negative and a particular product is a top topic, then you can immediately begin working on figuring out why the product is getting such bad commentary. Take a look at the CFC I attached to this entry. It is pretty simple and could probably be improved a bit. Consider it free and open source.
Archived Comments
Some reason they think that CFMumboJumbo.com talk about Sports and Women's clothing??? That's just wrong man.
Cool. :) Definitely going to give this a try myself.
I assume you'll be submitting your CFC wrapper to their Code Samples section (or already have?)
http://community.openamplif...
@Tim: Do you have ads? I do - it picked up on it I believe.
@Peter: Yep, emailed them yesterday. They responded today and I sent in my CFC and a simple CFM.
The real power comes from the ability to read and analyze press releases in realtime - stock trading: http://bit.ly/97inLB
That's a damn good example too Jim. Thanks.
@Ray, nope no adds. I am using Mura. The topicintentions Domains are sports / fishing fashion / womens clothing, maybe I don't have enough content for it to figure out, it claims I am gender neutral, quite distressing.
@Tim: Interesting. I got sports for mine too - but not at the top. I'd probably say ping the guys and provide feedback. In general though it seems to be mostly spot on.
Summary of article: Algos are already high frequency trading (HFT) the crap out of the stock while you're still on the word "The".
I thought we wanted negativity so we can get more page clicks for our ads? o_O
I tried the CFC after signing up for the service but I must be doing something wrong. I init the CFC with my api key and then call the parse function passing the root URL of my blog and all the resulting data is neutral or null. There are no topics listed, polarity is neutral, etc.
I changed the URL to your blog URL Ray and got the same results.
Well, the obvious next step is to simply look at the cfhttp result. Do a quick dump of it and tell me what you see.
I don't believe there's anything wrong with your CFC Ray. The request to their server is working and the CFHTTP results are coming back. I'm just seeing a bunch of neutral data elements which make me think they are the defaults. In other words, the bullet points of info you received such as you are male, highly decisive.... I'm getting nothing like that. I see empty or neutral data sets for TopicIntentions, Topic, Styles, etc. no matter what URL I use.
Makes me think their API is having issues or something.
Weird - I just ran my tests again and it is working ok. I guess hit up their contact form?
Good to know it's working for someone. I'll give it some time and check again in a few hours. If it doesn't work then I'll give them a shout. Thanks Ray.
I can just imagine what it'd report on my blog. hehehe.
Very intriguing indeed. And it seems that these folks are based not too far from me. Going to have to check them out . . .