I started searching for the most gender ambiguous names I could think of like "Jesse", "Alex", "Erin", etc. The best one I've found so far is "Angel" at 51.1% male.
why bother with a naive bayes classifier, why not just use the dictionary, when it matches the percentage is simple the ratio of the genders? I don't see the need for a classifier, I was hoping it was going to do something clever like guess the gender of a document's author.
While it's a quite interesting coding task to write a classifier, for the overwhelming majority of applications you simply don't need to know a user's gender. Making it a public API is a bad thing.
Developers have a horrible tendency to gather as much data on someone as possible, everything they're willing to give in fact, for the simple reason of "just in case we need it later". It's far, far better to gather as little as possible and build something that simply doesn't need to know specifics. If we build things that are ambiguous, unspecific for age, gender, race, nationality, etc then the world will be a better and more inclusive place. Paradoxical as it seems, more privacy actually leads to a more integrated society. That is universally a good thing (in my opinion, obv.).
Andrea is a female name in Spain, unless it's in very specific regions that I don't know of (source: I'm Spanish). I think you're right about Italy, though.
I'm Italian, my name is Gabriele and it classifies it as female. That's wrong. In Italy, Gabriele is a male name. I agree that this is useless without location information.
The only reliable way to find out someone's gender is to ask them what gender they currently think they are. There are more than two genders and an individual's gender can change with time.
A better strategy all-round would to be ask yourself whether you need to know someone's gender. I can think of very few legitimate reasons to know and record a user's gender and many of them can be dealt with by simply asking them what they'd like to be referred as, perhaps in more than one scenario.
One use I've come across for inferring genders was tackling record depupliction across multiple data sources. In one data set we might have the gender information of an individual but have it be missing in another.
Turns out gender is great to include in a blocking keys to reduce number of comparisons. Extrapolating an inferred gender in the dataset without one was incredibly helpful.
You are forgetting that the individual might consider the gender to be private information. Some people might want to use a different gender in different contexts. See the ESPN/Grantland suicide issue recently if you care/dare.
An individual that considers gender to be private information , and that uses different genders in different situations is very unlikely to be using a name that can be classified with a high degree of confidence as one gender.
A person using the name "Jack" is unlikely to be assumed to be a woman, even if that person selected "female" from a drop down somewhere. If the same person uses Jack/M and Cindy/F in different contexts, no fuzzy algorithm is going to resolve them as the same person (bar some other, stronger ID, such as a SSN).
EDIT: I initially used "William" as an example. Ironically, it turns out that name is only 57.6% male. Both Jack and Cindy are 90% male/female.
At least in the UK, Jack is a fairly common shortening of Jacqueline. The only way of determining a users gender is asking them directly, and if a person wishes to enter different values into different systems, all the more power to them.
Neat. Should probably call out that it's an API for "gender classification of english names". Did you build this mostly for learning/personal purposes?
Unlike a lot of people here, I don't think there is anything wrong with an API like this. It's true that it isn't culturally neutral, but there are times when any piece of information is useful.
I think it is important to make a distinction between gender and sex [1]. The link below is a makes a pretty good distinction between the two.
"Sex" refers to the biological and physiological characteristics that define men and women.
"Gender" refers to the socially constructed roles, behaviours, activities, and attributes that a given society considers appropriate for men and women.
45 comments
[ 3.2 ms ] story [ 99.7 ms ] thread[1]: http://en.wikipedia.org/wiki/Unisex_name
In the present case, the API is, unfortunately, incorrect.
Michael = 52%, Thomas = 62% (although Tom = 98%)
'Eric' (which often becomes Erica for females) scored a 10% higher chance of being male than 'John' did (Eric = 66.854585% while John = 57.399103%).
What's also surprising is the dramatic change the Anglicizing of the name brings: 'Erik' = 91.939346% while 'Eric' = 66.854585%
[1]: http://blog.xkcd.com/2014/01/31/the-baby-name-wizard/
[2]: http://gender.hankstoever.com/classify/Permelia
Developers have a horrible tendency to gather as much data on someone as possible, everything they're willing to give in fact, for the simple reason of "just in case we need it later". It's far, far better to gather as little as possible and build something that simply doesn't need to know specifics. If we build things that are ambiguous, unspecific for age, gender, race, nationality, etc then the world will be a better and more inclusive place. Paradoxical as it seems, more privacy actually leads to a more integrated society. That is universally a good thing (in my opinion, obv.).
Also this: http://www.cscyphers.com/blog/2012/06/28/falsehoods-programm...
A better strategy all-round would to be ask yourself whether you need to know someone's gender. I can think of very few legitimate reasons to know and record a user's gender and many of them can be dealt with by simply asking them what they'd like to be referred as, perhaps in more than one scenario.
Turns out gender is great to include in a blocking keys to reduce number of comparisons. Extrapolating an inferred gender in the dataset without one was incredibly helpful.
A person using the name "Jack" is unlikely to be assumed to be a woman, even if that person selected "female" from a drop down somewhere. If the same person uses Jack/M and Cindy/F in different contexts, no fuzzy algorithm is going to resolve them as the same person (bar some other, stronger ID, such as a SSN).
EDIT: I initially used "William" as an example. Ironically, it turns out that name is only 57.6% male. Both Jack and Cindy are 90% male/female.
You have just used the numbers from the API as evidence of its own accuracy.
http://en.wikipedia.org/wiki/Unisex_name has been mentioned elsewhere, but I've found it pretty useful.
Unlike a lot of people here, I don't think there is anything wrong with an API like this. It's true that it isn't culturally neutral, but there are times when any piece of information is useful.
"Sex" refers to the biological and physiological characteristics that define men and women.
"Gender" refers to the socially constructed roles, behaviours, activities, and attributes that a given society considers appropriate for men and women.
[1]- http://www.who.int/gender/whatisgender/en/
[1]:http://www.chw.org/display/PPF/DocID/22620/Nav/1/router.asp