Genderize.io determines the gender of a first name | Hacker News Reader

Genderize.io determines the gender of a first name

genderize.io

21 pointsduvander59 comments

Pro tip: Don't try to be "too smart" or clever when talking to your users.

Being wrong here is so much more tacky than either asking for gender if you need it, or just writing in a gender neutral way.

"I'm sorry we were wrong, but our computer determined you have a girl's name. If you change your name, it'll work now."

Yes. Don't attempt to infer gender from names.

Mostly, you don't need it. You'll often get it wrong. You may end up pissing off transgender people. Unless you are running a dating site (in which case, you should actually get people to specify which gender(s) they are—and, yes, people will want options beyond male and female) there is no need for it.

If you want to know my gender, ask. Don't infer.

I suspect this is more for data analysis than it is for user-facing features.

I hope.

This is far more useful for advertising, especially when trying to build user segments. If you have the gender from the user directly entering it somewhere along the line, great. If not, this comes in.

And in more cases than people seem to think, you really don't need it.

It seems to be broken, everything is returning male (or null)

    curl http://api.genderize.io?name=joe
    {"joe":"male"}

    curl http://api.genderize.io?name=jo
    {"jo":"male"}

    curl http://api.genderize.io?name=jamie
    {"jamie":"male"}

    curl http://api.genderize.io?name=jaime
    {"jaime":"male"}

    curl http://api.genderize.io?name=joan
    {"joan":"male"}

    curl http://api.genderize.io?name=kelly
    {"kelly":"male"}

    curl http://api.genderize.io?name=lucy
    {"lucy":null}

http://genderizer.herokuapp.com?name=joe

{"joe":"male"}

http://genderizer.herokuapp.com?name=jo

{"jo":"female"}

http://genderizer.herokuapp.com?name=jamie

{"jamie":"mostly_female"}

http://genderizer.herokuapp.com?name=jaime

{"jaime":"andy"}

http://genderizer.herokuapp.com?name=joan

{"joan":"female"}

http://genderizer.herokuapp.com?name=kelly

{"kelly":"mostly_female"}

http://genderizer.herokuapp.com?name=lucy

{"lucy":"female"}

BONUS: http://genderizer.herokuapp.com/?name=jaime&country=ireland

{"jaime":"male"}

Heroku's free tier, not nearly as overloaded right now, but still. Source here: https://github.com/rajington/genderizer

Very broken. Many common male names are returning null as well:

curl http://api.genderize.io?name=sean {"sean":null}

http://genderizer.herokuapp.com/?name=sean {"sean":"male"}

Heroku's free tier, not nearly as overloaded right now, but still. Source here: https://github.com/rajington/genderizer

It's intensely slow too. I just fed it top 10 female names of the 70s and 80s and it got about 30%. It's wrong, slow, and broken. Not a good start.

  ab -n 100 -c 5 http://api.genderize.io/\?name\=Cameron

Total: 636 1373 157.7 1304 1664

  ab -n 100 -c 5 http://genderizer.herokuapp.com/\?name\=Cameron

Total: 77 141 70.1 115 383

Heroku's free tier, not nearly as overloaded right now, but still. Source here: https://github.com/rajington/genderizer

For people interested in other tools, there is a command line gender verification tool and database:

http://www.autohotkey.com/board/topic/20260-gender-verificat...

and a Ruby gam that is built on the above data:

https://github.com/bmuller/sexmachine

  http://api.genderize.io/?name=jessie
  {"jessie":"male"}

  irb(main):005:0> d.get_gender("Jessie")
  => :female

Not only is it better, it handles cases like androgynous and "mostly male"

BRB while I turn this into a web api

Done https://github.com/rajington/genderizer

and Done

http://murmuring-depths-3923.herokuapp.com/?name=jessie

http://murmuring-depths-3923.herokuapp.com/?name[0]=peter&na...

Compared to:

http://api.genderize.io/?name=jessie

http://api.genderize.io/?name[0]=peter&name[1]=lois&name[2]=...

I just wanted to learn basic Sinatra, but I might try to do the POST operations too.

Finished POST, changed the domain to genderizer.herokuapp.com:

http://genderizer.herokuapp.com/?name=jessie

http://genderizer.herokuapp.com/?name[0]=peter&name[1]=lois&...

A "command line gender verification tool"? So it'll verify my gender based from the command line? :P

I ran it...

....I....I.... just want to be alone for a bit.

Another DB: http://svn.php.net/viewvc/pecl/gender/trunk/data/nam_dict.tx...

Neat!

It would be nice to get some sort of "confidence level" - http://api.genderize.io/?name=jackie

http://genderizer.herokuapp.com/?name=jackie {"jackie":"andy"}

Source here: https://github.com/rajington/genderizer

My first test http://api.genderize.io/?name=Erin gives me "Male"

http://genderizer.herokuapp.com/?name=Erin {"Erin":"mostly_female"}

Source here: https://github.com/rajington/genderizer

A "confidence level" will now be recieved :)

It's strange to believe you can effectively determine gender from a first name - even when concentrating on Anglo-centric names, as this appears to.

How is the data generated? By hand? I can imagine this might be somewhat useful if it gave a non-binary result.

"Paris" is "male".

Paris is "andy" androgynous

http://genderizer.herokuapp.com/?name=Paris

Source: https://github.com/rajington/genderizer

Interesting, but I question its utility, particularly if a name is ambiguous - like "Alex". Also, I tried it with my sister's name, Nanna, and it returned "null." Apparently she's neuter.

On my first try I got an erroneous result on a non-Anglo name: http://api.genderize.io/?name=krishna returned "male". In fact god names are unisex; I've met a female Krishna.

Edit: and "null" on Vishnu. I didn't see on the page whether this is supposed to extend beyond Anglo-American culture.

And "kyle" came up "male", tho I can report a female Kyle. Whoever is populating the database may be underestimating gender/naming trends.

Other commenters are already demonstrating that the error rate is way too high to be useful. If you need a user's gender (and you probably don't), just ask them.

Handy for ads.

It seems like he deleted my comment from his site's disqus (it's his right) but I rewrote the exact same API in a couple of lines using sexmachine and ruby (I had been meaning to try Sinatra anyway). The source is here: https://github.com/rajington/genderizer

Hosted on heroku's free tier here are some more interesting examples:

Basic: http://genderizer.herokuapp.com/?name=peter

  {"peter":"male"}

"Mostly" Response: http://genderizer.herokuapp.com/?name=Jamie

  {"Jamie":"mostly_female"}

Country (helps with guesses, default USA): http://genderizer.herokuapp.com/?name=Jamie&country=great_br...

  {"Jamie":"mostly_male"}

I've thought about this before, you also should be able to get a somewhat accurate estimate of the median age of a group of attendees to an event based on the first names alone. See Baby Name Wizard Voyager tool for visualizations of first name occurences since 1900s: http://www.babynamewizard.com/voyager

I would be impressed by that!

Hello.

I'm the author of genderize.io. At the time of your comments i was slowly getting this API up and running. A lot of sudden traffic made that a little hard :)

At the time, i was utilizing a few lists of officially approved names from a few European countries. We have those in Denmark, where i'm from, among a few other countries. This had led me to believe that most countries had lists like this and i was planning to add them as i went along.

I since change my approach, moving from the lists to large datasets of user profiles from social networks. This has given me the chance to update the API, adding probabilistic guesses along with the ability to add localization based filters.

My datasets are not that huge yet, but they're constantly growing.

Just wanted to say thanks for all the feedback i found in here and give a heads up on the update :)

There has been a perl module that does this for a while http://search.cpan.org/~edaly/Text-GenderFromName-0.32/Gende... I see you can add names to its "dictionary" and it give a probability.

First two tries:

  ~ % curl "http://api.genderize.io?name=pat"
  {"pat":null}%

Shouldn't that be "unisex"?

  ~ % curl "http://api.genderize.io?name=patricia"
  {"patricia":null}%

Hm...

Completely worthless without a national/ethnic hint.

E.g. "Jean" will be male if the first name of a French person, while likely female in the US.

As Elizabeth or Leslie are common male names in the UK but female in the US.

I appreciate the authors effort, but don't see much use.

Never heard of a male "Elizabeth" here in the UK

Elizabeth isn't used as a male name anywhere...

It tells me that my daughter's name, Avery, is male, yet it is the 13th most popular name for baby girls right now. Someone needs an updated data set.

How well does it compare to https://nltk.googlecode.com/svn/trunk/doc/book/ch06.html. If the genderize.io folks are listening, can you look at providing some benchmark/test set to evaluate the efficiency of the underlying algorithm?

I haven't tried the api yet, but the example on the home page says that "Lois" is a female name - when in fact, in the 1930s it was quite a popular male name with about 70 boys born each year named lois. Checkout http://www.namemeababy.com to see the full stats.

Worked on gender-determination for a project for a client in a U.S. male-dominated industry. Baby name gender frequencies by birth year are available online. One interesting wrinkle - can apply simple Bayesian technique - e.g. even if Pat is 50-50 male-female in population at large, firefighter Pat is more likely to be a man.

If you're interested in doing this yourself, the book associated with Python's Natural Language Toolkit has a pretty good introduction to identifying gender in the chapter on text classification [1].

[1] http://nltk.org/book/ch06.html

{"ashley":"male"}

Also helpful is:

https://github.com/bmuller/sexmachine

which uses a multilingual dictionary, that accounts for names that cross gender in different countries.

The underlying dictionary for this, gender/nam_dict by Jorg Michael, is an awesome piece of work: it has entries for each name, its gender (or genders), and how popular that name is in each country where it's used (especially useful for names which are female in one country but male in another country, but where they're much more popular in the former than the latter).

It's a bit of an odd file format but it's fairly straight forward to build something which will read it into the language of your choice.

http://www.autohotkey.com/community/viewtopic.php?t=22000

I sanity-checked the service with some common unisex names. Here are the results I got:

Unisex: Robin, Alex

Female: Kelsey, Carol

Male: Leslie, Jodi, Jody, Frankie, Kelly, Marion, Casey, Jamie, Jordan, Mackenzie

Null: Pat, Taylor, Dale

Hey Ryan

I think most of these names will result in prober responses now.

If this were for data data analysis then it would have a probability attached.What you have to ask yourself is gender really needed for your application? If not it's best to just avoid it; using guesses is the best way to data corruption.

There's a PECL extension for PHP that claims to provide this too: http://php.net/gender

Not tested it on too many challenging tasks but it seems to work well too.

http://api.genderize.io/?name=Αλεξανδρος

This throws an error and gives us a very pretty stack-trace :)

I couldn't find an API, but some might be interested in GenderChecker [0].

[0] http://www.genderchecker.com

Goes to a synology login page: http://www.genderize.io

Apparently I'm "null". I'll assume that's not an attempt to be deliberately offensive. :-)

I wonder what dataset is being used behind the scenes. Tried "Roberta" and it returned "null".

http://api.genderize.io/?name=pat

This sort of thing sounds like something Facebook would be very good at doing using Open Graph.

How does that site work? It mentions nothing about pricing?

Reply on news.ycombinator.com