happy factor
Two friends and I are working on a fun side project called HappyFactor where we randomly send text messages to people throughout the day and ask them how happy they are, and what they are doing.
The idea is threefold. First, just by having people think about their happiness in their current context, they can develop mindfulness and increased self-awareness. Look down from the TV screen, and get in touch with your emotions. Are you really happy? If so, is it the people around you that are making you happy? Other emotional states…stress? Boredom? Euphoria?
Second, by logging in to the web site, you get feedback about your history. This is cool because you can see graphs of your happiness history, when you were happiest, what days of the week, times of the day. Human memory sucks, that’s why random texting is the bomb for figuring out, objectively, what you were thinking and doing in the past. The hippocampus has nothing on my Poisson process!!! It does have flaws…for example, we can’t capture dreams. But just like Google is slowly replacing my semantic memory, we can start to outsource episodic memory to HappyFactor. Then brains can work on more important stuff, like labeling data.
Aside: Thanks to operant conditioning, the physical sensation of my cell phone buzz now prompts me to stop and think about myself, even if the text turns out to be from someone else. I’m ratcheted up to an average of four texts per day and the effect is pretty reliable.
Third, once we have everyone texting their happiness all the time, we can do controlled experiments in the population to find predictive features of population happiness. Like give Cincinnati “Free Cone Day” and nothing to Cleveland. You could even do it district by district, assuming we had enough people. This would give us a good predictive model of population happiness. Then at a really macro level we could have a big model for the entire U.S. population and optimize the shit out of it. Bhutan does something like this already.
Recently I started working on a feature that would allow you to see words correlated with different happiness levels. I tried out MontyLingua because I wanted a simple POS tagger in Python that I could mod out. It doesn’t work that well with the types of sentences we have, which are generally sentence fragments (a typical message might be “6 at work”), although to be fair I haven’t played around with it a whole lot. The noun phrase and verb phrase hit rate is pretty low.
For now, I just use the stemmer (“data” -> “datum”, woo!) and a white list of stop words and do single word frequencies. Lots of false positives, the subtle nuance of a “phrase” is missed…but the recall is dece, and that’s what’s most important. Ideally I would show one chart for nouns and one for verbs. The most obvious application is to tell HF who you are with every time you get texted, and let the application figure out whom you should spend more time with.
Words are ranked from happiest to unhappiest, colored from blue to purple to red, and sized according to frequency. The Lovefest dance music festival in San Francisco is my favorite day of the year, which the data bears out.
3 years ago