There is a site GenderAnalyzer.com that claims to be able to determine the gender of a web page author. I was somewhat sceptical to begin with but on December 1 I entered the URL for this blog http://blog.lightingonemorecandle.com/ and the site told me that they were 61% sure it was written by a woman. Of course they were wrong; all of it was written by me. There was a check box to indicate if they were right or wrong so I checked Wrong and hit Submit and the next page showed the results of 802 Correct and 671 Incorrect. So they were correct 54% and incorrect 46% of the time.

So not being able to fall asleep I tried it again at a bit after midnight on December 5 and this time it said that it was 62% sure that this blog was written by a woman. Of course it was wrong and what is more interesting is that there was not change or new posts or comments made between the two tests. Now they claim that they are at 1150 Correct and 965 Incorrect which is still in the 54% to 46% rate.

I was very sceptical of this Gender Analysis program but I thought it might be amusing. Well it was amusing enough to be the subject of this blog post but not much more. If a person is setting up a system to do any kind of analysis it is usually a good idea to the the system store knowledge about the specific input. So ideally it would have been slightly more reasonable if the analysis program had stored the URL, the dates of posts and other information about the page along with the crucial piece of information the this blog is written by a man. I could have then reported two values; what it found based on prior knowledge and what it computed based on its algorithms.

No comments: