I agree with you. In addition, they forgot to consider Twitter's use as a marketing tool for small businesses. Typically, small businesses tend to use happy words and phrases to describe their products. In addition, social networks are "generally" happy places. I don't think I have come across individuals using social networks as places to vent their sadness for long....
On a similar note, one must wonder to what extent those who tweet are even local. For Napa, CA, which is mentioned in a comment, that would probably not be the case. Another example is Nevada, which, according to their map, must be something like the second happiest state in the union. I live in Nevada, and I am indeed very happy, but that's because I live in Tahoe, a rather atypical corner of the state. For Nevada as a whole, I am sure the result is due to things like tourists in Vegas going ga-ga over all Das Blinkenlighten, or gamblers tweeting when they win but not when they lose, the latter being the default.
It would be interesting to see how the results would differ were this method used with groups of adjacent words instead of just individual words.
Analysis based on idividual words might miss a lot of subtle context, especially since English is such an idiomatic language. For instance, focusing on individual words (or even small groups of adjacent words) may lead to mistaking a morbid joke for sadness, while a sarcastic insult might be mistaken for happiness.
Still, this is an ingenious way to measure happiness, and I'm looking forward to reading future research that builds on this work.
> we avoid stemming words, i.e., conflating inflected words with their root form, such as all conjugations of a speci c verb. For verbs in particular, by focusing on the most fre- quent words, we obtained scores for those conjuga- tions likely to appear in texts, obviating any need for stemming. Moreover, while we observe stemming works well in some cases for happiness measures, e.g., havg(advance)=6.58, havg(advanced)=6.58, and havg(advances)=6.24, it fails badly in others, e.g., havg(have)=5.82 and havg(had)=4.74; havg(arm)=5.50 and havg(armed)=3.84; and havg(capture)=4.18 and havg(captured)=3.22.
That is something I came across myself when I was working on a similar data mining project. But this was a surprise:
> In terms of methodology, our hedonometer could be improved by incorporating happiness estimates for common n-grams, e.g., 2-grams such as `child abuse' and `sex scandal' as well as negated sentiments such as `not happy'
I can understand ignoring n-grams but a simple no-X or not-X is a necessity. In common parlance, 'Peace' > 'No war' > 'No peace' > 'War'. If you assume the simple happiness values of 'Peace' = 1, 'War' = -1, 'No' = -0.5, then the scores end up: Peace=1, No War=-1.5, No peace=0.5, War=-1. Clearly 'War' is worse than 'No War' yet the scores don't correspond. The correct order should be something like: Peace=1, No War=0.5, No peace=-0.5, War=-1. To make this work, 'No' shouldn't be a fixed negative value like -0.5. It should be -(next-word/2). Then 'No Peace' = -0.5 + 1 = 0.5 and 'No War' = 0.5 + -1 = -0.5.
That's funny because when I saw the title of this thread, my gut said "San Luis Obispo, surely!" I've only visited a few times but the people seem so genuinely pleasant and the big things all seem just right (enough so that the property prices are through the roof, naturally.. so maybe it's not perfect ;-))
Meh... I can't really put my finger on it, but something just seems off (or unrigorous) with this. I feel like if I wanted to come up with some kind of metric to demonstrate that southern or northern states are the happiest (over western), I probably could.
In other words, start with my results and then find some data to support them.
Interesting, but bullshit as titled and described.
Regional happiness may or may not correlate to usage of these "happy words" or "unhappy words", especially across local cultures, circumstances (e.g. before or after Obama's election, different regions will respond differently), technology/twitter adoption, etc.
I'm surprised New Yorkers aren't the bunch of miserable bastards I thought we were. Also I realise money may not buy you happiness, but a lot of the wealthier areas tended to show more happiness, although that may just be a function of poverty creating sadness.
I can't imagine “restaurant”, “wine”, and even “cheers” going with people who can't afford any.
Judging positive/negative emotions by semantic analysis of 140 chains of characters. Meh ? Call me unconvinced. Especially for people who tend to use irony in their communications, semantic analysis can be easily misinterpreted. Plus, who says that the people tweeting in a particular place are actually living there and not just passing by ? Who says their tweets relate to their environment and not to a particular news they reacted to ?
I think it can certainly be used as a "qualitative" analysis, but making it a "measurement of happiness" is overdoing it.
"With a score of 6.25, we found the happiest city to be Napa, CA, due to a relative abundance of such happy words as “restaurant”, “wine”, and even “cheers”, along with a lack of profanity."
I grew up in Napa so I know something was off here. While adults love it, most kids find it boring or even hate growing up there.
Tweets about wine != happiness. Lack of profanity != happiness. Examples - "The wine sucked" or "Vegas is f*cking awesome!" This is an interesting study of vocabulary, but hardly a measure of happiness.
You do not necessarily need a perfect 1-on-1 relationship between the used words and 'happiness', as long as this 'noise' is canceled out in the same way between areas.
And that's where there might be a problem, the results may (perhaps partially) reflect a structural difference of 'use of language' between areas.
From my experience of living in lots of places, I'd say that methodology that finds New York and Washington state as happy and Texas and Louisiana as sad is fundamentally flawed.
Attempting to use tweeted words as indicators of happiness ignores context.
28 comments
[ 3.7 ms ] story [ 80.3 ms ] threadAnalysis based on idividual words might miss a lot of subtle context, especially since English is such an idiomatic language. For instance, focusing on individual words (or even small groups of adjacent words) may lead to mistaking a morbid joke for sadness, while a sarcastic insult might be mistaken for happiness.
Still, this is an ingenious way to measure happiness, and I'm looking forward to reading future research that builds on this work.
> we avoid stemming words, i.e., conflating inflected words with their root form, such as all conjugations of a speci c verb. For verbs in particular, by focusing on the most fre- quent words, we obtained scores for those conjuga- tions likely to appear in texts, obviating any need for stemming. Moreover, while we observe stemming works well in some cases for happiness measures, e.g., havg(advance)=6.58, havg(advanced)=6.58, and havg(advances)=6.24, it fails badly in others, e.g., havg(have)=5.82 and havg(had)=4.74; havg(arm)=5.50 and havg(armed)=3.84; and havg(capture)=4.18 and havg(captured)=3.22.
That is something I came across myself when I was working on a similar data mining project. But this was a surprise:
> In terms of methodology, our hedonometer could be improved by incorporating happiness estimates for common n-grams, e.g., 2-grams such as `child abuse' and `sex scandal' as well as negated sentiments such as `not happy'
I can understand ignoring n-grams but a simple no-X or not-X is a necessity. In common parlance, 'Peace' > 'No war' > 'No peace' > 'War'. If you assume the simple happiness values of 'Peace' = 1, 'War' = -1, 'No' = -0.5, then the scores end up: Peace=1, No War=-1.5, No peace=0.5, War=-1. Clearly 'War' is worse than 'No War' yet the scores don't correspond. The correct order should be something like: Peace=1, No War=0.5, No peace=-0.5, War=-1. To make this work, 'No' shouldn't be a fixed negative value like -0.5. It should be -(next-word/2). Then 'No Peace' = -0.5 + 1 = 0.5 and 'No War' = 0.5 + -1 = -0.5.
PS: I typed 'War' so many times in the above sentence that I had to check its spelling to confirm it was a real word. http://en.wikipedia.org/wiki/Semantic_satiation strikes again.
I didn't see it on the list here, I'm assuming because the population was too small.
http://travel.usatoday.com/destinations/story/2011/04/San-Lu...
How the heck did you get that domain?
In other words, start with my results and then find some data to support them.
Regional happiness may or may not correlate to usage of these "happy words" or "unhappy words", especially across local cultures, circumstances (e.g. before or after Obama's election, different regions will respond differently), technology/twitter adoption, etc.
I can't imagine “restaurant”, “wine”, and even “cheers” going with people who can't afford any.
I think it can certainly be used as a "qualitative" analysis, but making it a "measurement of happiness" is overdoing it.
I grew up in Napa so I know something was off here. While adults love it, most kids find it boring or even hate growing up there.
Tweets about wine != happiness. Lack of profanity != happiness. Examples - "The wine sucked" or "Vegas is f*cking awesome!" This is an interesting study of vocabulary, but hardly a measure of happiness.
And that's where there might be a problem, the results may (perhaps partially) reflect a structural difference of 'use of language' between areas.
Attempting to use tweeted words as indicators of happiness ignores context.