Why does a Google search for unprofessional hair show images of black women?(iafrikan.com)
iafrikan.com
Why does a Google search for unprofessional hair show images of black women?
http://www.iafrikan.com/2016/06/25/why-does-a-google-search-for-unprofessional-hair-show-images-of-black-women-including-michelle-obama-2/
3 comments
>> According to Google, "[Image Search] analyzes the text on the page adjacent to the image, the image caption and dozens of other factors to determine the image content."
The reference for this (missing from the op) seems to be a paper titled "Evolving data mining into solutions for insights" [1]. I don't have time to read this now and the one-liner abstract is not very informative.
Still, it dates from 2002, whereas allegedly Google Image Search nowadays uses actual image processing. There was a different article submitted on HN earlier today that basically says Google's machine learning algorithms used in Image Search now "understand" what the image content is [2] (which is patently b.ll.cks but, hey).
My best guess is that Image Search does use more modern machine learning but still operates largely in the old model, where it looks primarily in the text accompanying an image- and in fact even treats that text as a good, old-fashioned "continuous bag of words" rather than more modern techniques like word embeddings, or even N-grams.
Edit: basically what I'm saying is that Google seems to take in a query like "unprofessional hair" and return all documents that have the words "unprofessional" and "hair" in them, even when "unprofessional" is qualified with "not" very clearly (the retrieved documents are tiny).
That seems a bit weird- even Google web search seems to be better than that at taking some context into account.
[1] http://dl.acm.org/citation.cfm?doid=545151.545174
[2] https://news.ycombinator.com/item?id=11975878
The reference for this (missing from the op) seems to be a paper titled "Evolving data mining into solutions for insights" [1]. I don't have time to read this now and the one-liner abstract is not very informative.
Still, it dates from 2002, whereas allegedly Google Image Search nowadays uses actual image processing. There was a different article submitted on HN earlier today that basically says Google's machine learning algorithms used in Image Search now "understand" what the image content is [2] (which is patently b.ll.cks but, hey).
My best guess is that Image Search does use more modern machine learning but still operates largely in the old model, where it looks primarily in the text accompanying an image- and in fact even treats that text as a good, old-fashioned "continuous bag of words" rather than more modern techniques like word embeddings, or even N-grams.
Edit: basically what I'm saying is that Google seems to take in a query like "unprofessional hair" and return all documents that have the words "unprofessional" and "hair" in them, even when "unprofessional" is qualified with "not" very clearly (the retrieved documents are tiny).
That seems a bit weird- even Google web search seems to be better than that at taking some context into account.
[1] http://dl.acm.org/citation.cfm?doid=545151.545174
[2] https://news.ycombinator.com/item?id=11975878
meira(2)
They were talking about how it is now largely an Hispanic area. They talked to a young black girl, around age 9:
--- begin quote ---
When JaTiara Fuller, walked into the classroom on her first day of fourth grade, she was surprised. She was the only black person in her class. “There was no diversity,” she remembers. “There were no Indians, no Asians, no Filipinos, just Latinas and me. That’s it.”
Fuller's mom, who grew up in Compton, was sure the school had made a mistake. She thought her daughter had been put in a class for students who needed to learn English, so she went to the administrator’s office to sort things out. “And they were like, no she’s in the gifted class, that’s where she is supposed to be,” JaTiara says.
She made friends easily with other kids, though there were a few things that made her stand out. "They couldn't relate to my hair problems or my butt problems”, Fuller says. “It was a foreign world to them.”
--- end quote ---
So, if black women having hair problems is indeed a big thing among black women, then it would not be surprising that Google knows it.
[1] http://www.marketplace.org/topics/wealth-poverty/straight-ou...