[jboons]

My initial impression is it doesn't seem analyze groups of words very well. It will pick emotional words out of the middle of a sentence and apply the emotion to the whole text, when the context of the sentence changes the meaning of the word.

Quick example, "I hate your guts and I hope you die." -- Hope is picked out and noted as cheerful.

[kristopolous]

my frustrometer project, which does the same thing and predates this by a few years (albeit less flashy), at http://frustrometer.com/ can handle that fine.

I didn't finish all the build out, because I got really cold feet in the early pitches ... moved on to other things

should I get back on this? It's always so hard to tell if discouragement is a true representation or just people being internety. I also grossly lack self confidence. I don't know if I need pills, therapy, or what.

[arcameron]

Keep building it. Especially if it interests you. It interested you enough to start it in the first place! Just don't build it with the expectation that you'll garner fame or fortune.

I imagine a lot of people could use some kind of OSS sentiment detection, even if it's not perfect.

Try therapy before pills

[sp332]

I think it's aimed at business communication. It picks out terms that might be taken the wrong way, instead of trying to rewrite your whole communication.

[jsprogrammer]

So that your layoff notice can be 100% 'agreeable'?

[kristopolous]

when I was doing my stab at this problem (see comment above), the contextual power structure was important.

that's why I coupled my model with flow analysis on the mail server end to assign subordination relations between parties when possible.

A general solution is OK for liaisons but that's about it.

The IBM demo appears to be just positioning Watson as a general business concierge here and doesn't seem to be concerned with a true assist on what Gardener calls interpersonal intelligence.

Being able to write concisely and well doesn't necessarily mean you can permeate communication walls.

A valid computer solution, in brief, involves continuous kernel application during the authorship; how the user responds to the system is itself part of the analysis. This isn't just binary sentiment analysis of product reviews using an SVM - that's childs' play compared to this.

Even with the best solution I came up with, it's still just culturally specific ... cultural to the geolocation of the persons and the heritage of the people, but also to the industry itself.

Something that may show weakness in one industry may show wisdom in another, hostility in another, and humility in another.

For instance, say you met someone with whom you had a serious romantic interest. Would you exchange business cards, and then arrange a 9am skype call the following tuesday? It's just all so contextual.

This rabbit hole runs deep.

[jsprogrammer]

? Which project?

I'd be interested in whatever else you can talk about. I find this topic interesting and can see the usefulness of it (though I'm imagining something more comprehensive than this demo). The kind of nightmare that I imagined when going into this article though, was some kind of genetic/optimization algorithm that will just iterate on some text until it has achieved the optimal, "most emotionally/etc correct", communication. Obviously, this assumes some highly accurate model of people's emotional perceptions [not to mention, what the words actually mean :)], but I think it's still a somewhat scary thought.

Anyway, would love to hear more about what you're working on.

[kristopolous]

I have an open source version, but it's intentionally about 800 commits behind my current work (which I haven't touched in a while tbh). If you are really interested, let's take this offline. kristopolous (at) gmail.com

[wbhart]

I tried just that: "your work is simply not engendering confidence. If I were working for a sewerage works I'd recommend you for promotion, since what you are producing is more of the proverbial. By Friday things will have improved one way or another. I am sure you understand what I mean." The particular things that Watson saw in this just aren't what I see in it.