General Comments/Tracking

[jessica-writes-code]

Gender Analysis:

• https://genderize.io/#overview

Literature:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4740422/

[shazad]

Literature:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4740422/ (good overview of methodology employed)
• http://www.aaai.org/ocs/index.php/ICWSM/ICWSM15/paper/view/10585/10528
• http://asr.sagepub.com/content/80/5/960.full
• http://www.tandfonline.com/doi/full/10.1080/1461670X.2013.834149?scroll=top&needAccess=true (includes a long term analysis on gender in media going back to 1880 for the NYT)
• http://flaounas.com/images/FLAOUNAS_2011__THESIS.pdf
• http://mediapatterns.enm.bris.ac.uk/pdf/Information%20Fusion%20for%20Entity%20Matching%20in%20Unstructured%20Data.pdf
• http://asr.sagepub.com/content/80/5/960.full (they use the Lydia news analysis system to identify the gender of individuals named in news reports)
• http://onlinelibrary.wiley.com/doi/10.1111/ssqu.12091/full (potential methods for gender identification in data and analysis section)

[jessica-writes-code]

Word2Vec

• Overview: https://arxiv.org/pdf/1301.3781.pdf; https://blog.acolyer.org/2016/04/21/the-amazing-power-of-word-vectors/; https://arxiv.org/pdf/1301.3781.pdf; https://arxiv.org/abs/1411.2738
• https://www.researchgate.net/publication/305615978_Man_is_to_Computer_Programmer_as_Woman_is_to_Homemaker_Debiasing_Word_Embeddings (Saligrama)
• https://levyomer.files.wordpress.com/2014/04/linguistic-regularities-in-sparse-and-explicit-word-representations-conll-2014.pdf
• https://arxiv.org/pdf/1606.06121v1.pdf
• How-To: https://rare-technologies.com/word2vec-tutorial/

Possible Uses of Word2Vec

• Word closeness (e.g., he is close to X, Y, Z, and she is close to A, B, C)
• Analogy generation (per Saligrama, top of pg. 8)
• Projection of words onto "Gender-Axis"

Heistations Re: Word2Vec

• It's tricky to explain

Other Stuff: Word Count

• http://www.tandfonline.com/doi/pdf/10.1080/00220480109595184?needAccess=true (Race, Ethnicity, Gender representation in economics text books)

Industries

• Business: http://www.tandfonline.com/doi/pdf/10.1080/08985629700000013; http://www.tandfonline.com/doi/abs/10.1080/08985626.2010.520338
• Science: http://pus.sagepub.com/content/early/2009/04/07/0963662508098580.abstract

[poorna-kumar]

Minutes of meeting 11/10:

• Social science question:
  1. Consensus is to go with our proposal (studying the representation of women in different industries). Studying across industries, and time, how representations of women change is an important research question in itself.
  2. In addition, does the authorship of an article affect its content (apply same methods we used for point 1, but now separate article groups into sub-groups of articles written by men and those written by women).
  3. Finally, we could contrast time trends of our results with the actual participation of women in those industries over time.
• Methods:
  1. Word counts.
  2. Similarity scores.
  3. Networks.
• Work plan until Tuesday 11/15:
  1. Have a crude network, and know whether this is something we should pursue. (Poorna)
  2. Work on identification of gender. (Poorna)
  3. Word counts -- figure out what/how. (Binghai)
  4. Similarity scores -- build yearly subsets, etc. (Jessica)
  5. Social science awesomeness. (Shazad)

[shazad]

Topics: Art Business Crime Elections Entertainment Fashion Politics Religion Science Sports Technology Weather Other

[jessica-writes-code]

@poorna-kumar

Hi Poorna -

Are the gender lists for each year basically stable? I was going to go ahead and start making author-gender-specific word vectors.

• Jessica

[poorna-kumar]

Hi Jessica,

So I think they are stable, except in the case of multiple writers, because those cases are not being handled correctly. I'm not sure, however, that we should be creating author-gender word vectors, primarily because different authors write about different topics, so the effect of the particular topic that is being covered will be very pronounced, and the results will be more telling of the topic being covered than the author. It might be hard to find authors of different genders who write about the exact same thing, so it might be hard to compare our results meaningfully. Aggregation seems preferable to me, the way you've already done.

On Nov 27, 2016 2:55 PM, "Jessica" notifications@github.com wrote:

@poorna-kumar https://github.com/poorna-kumar

Hi Poorna -

Are the gender lists for each year basically stable? I was going to go ahead and start making author-gender-specific word vectors.

• Jessica

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1#issuecomment-263154071, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_bzM3egawyyAkgWmpSBBiAlm4258Zks5rCgp3gaJpZM4KeWVh .

[jessica-writes-code]

That seems reasonable to me. I suppose I didn't realize the extent to which different genders of author cover different topics.

I'll remove/comment out the author-gender analysis.

On Mon, Nov 28, 2016 at 9:48 AM, poorna-kumar notifications@github.com wrote:

Hi Jessica,

So I think they are stable, except in the case of multiple writers, because those cases are not being handled correctly. I'm not sure, however, that we should be creating author-gender word vectors, primarily because different authors write about different topics, so the effect of the particular topic that is being covered will be very pronounced, and the results will be more telling of the topic being covered than the author. It might be hard to find authors of different genders who write about the exact same thing, so it might be hard to compare our results meaningfully. Aggregation seems preferable to me, the way you've already done.

On Nov 27, 2016 2:55 PM, "Jessica" notifications@github.com wrote:

@poorna-kumar https://github.com/poorna-kumar

Hi Poorna -

Are the gender lists for each year basically stable? I was going to go ahead and start making author-gender-specific word vectors.

• Jessica

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263154071, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_ bzM3egawyyAkgWmpSBBiAlm4258Zks5rCgp3gaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1#issuecomment-263341484, or mute the thread https://github.com/notifications/unsubscribe-auth/AKKw-nsBGpQnbaPYET7a229FnyI0mw_Nks5rCxPwgaJpZM4KeWVh .

[poorna-kumar]

Hmm, it's possible I didn't understand what you meant. We can conference in class. I thought by author-gender, you meant some cross product of author and gender. If you're picking all articles written by male/female authors, and finding word vectors for those, though, it should be okay, according to me.

On Nov 28, 2016 9:50 AM, "Jessica" notifications@github.com wrote:

That seems reasonable to me. I suppose I didn't realize the extent to which different genders of author cover different topics.

I'll remove/comment out the author-gender analysis.

On Mon, Nov 28, 2016 at 9:48 AM, poorna-kumar notifications@github.com wrote:

Hi Jessica,

So I think they are stable, except in the case of multiple writers, because those cases are not being handled correctly. I'm not sure, however, that we should be creating author-gender word vectors, primarily because different authors write about different topics, so the effect of the particular topic that is being covered will be very pronounced, and the results will be more telling of the topic being covered than the author. It might be hard to find authors of different genders who write about the exact same thing, so it might be hard to compare our results meaningfully. Aggregation seems preferable to me, the way you've already done.

On Nov 27, 2016 2:55 PM, "Jessica" notifications@github.com wrote:

@poorna-kumar https://github.com/poorna-kumar

Hi Poorna -

Are the gender lists for each year basically stable? I was going to go ahead and start making author-gender-specific word vectors.

• Jessica

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263154071, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_ bzM3egawyyAkgWmpSBBiAlm4258Zks5rCgp3gaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263341484, or mute the thread https://github.com/notifications/unsubscribe-auth/AKKw- nsBGpQnbaPYET7a229FnyI0mw_Nks5rCxPwgaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1#issuecomment-263341970, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_b5QVKEOFBVvkicW6bwtwmDabKuMjks5rCxRagaJpZM4KeWVh .

[jessica-writes-code]

Ah. No. Sorry. I obviously was not sufficiently clear.

I meant that I was taking articles from all male authors (in a year) and creating one set of word vectors and then taking articles from all female authors (in a year) and creating another set of word vectors.

As you said, we can conference in class. But I think that we're on the same page.

On Mon, Nov 28, 2016 at 10:15 AM, poorna-kumar notifications@github.com wrote:

Hmm, it's possible I didn't understand what you meant. We can conference in class. I thought by author-gender, you meant some cross product of author and gender. If you're picking all articles written by male/female authors, and finding word vectors for those, though, it should be okay, according to me.

On Nov 28, 2016 9:50 AM, "Jessica" notifications@github.com wrote:

That seems reasonable to me. I suppose I didn't realize the extent to which different genders of author cover different topics.

I'll remove/comment out the author-gender analysis.

On Mon, Nov 28, 2016 at 9:48 AM, poorna-kumar notifications@github.com wrote:

Hi Jessica,

So I think they are stable, except in the case of multiple writers, because those cases are not being handled correctly. I'm not sure, however, that we should be creating author-gender word vectors, primarily because different authors write about different topics, so the effect of the particular topic that is being covered will be very pronounced, and the results will be more telling of the topic being covered than the author. It might be hard to find authors of different genders who write about the exact same thing, so it might be hard to compare our results meaningfully. Aggregation seems preferable to me, the way you've already done.

On Nov 27, 2016 2:55 PM, "Jessica" notifications@github.com wrote:

@poorna-kumar https://github.com/poorna-kumar

Hi Poorna -

Are the gender lists for each year basically stable? I was going to go ahead and start making author-gender-specific word vectors.

• Jessica

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263154071, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_ bzM3egawyyAkgWmpSBBiAlm4258Zks5rCgp3gaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263341484, or mute the thread https://github.com/notifications/unsubscribe-auth/AKKw- nsBGpQnbaPYET7a229FnyI0mw_Nks5rCxPwgaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1# issuecomment-263341970, or mute the thread https://github.com/notifications/unsubscribe-auth/ASd_ b5QVKEOFBVvkicW6bwtwmDabKuMjks5rCxRagaJpZM4KeWVh .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poorna-kumar/MSE231-Project/issues/1#issuecomment-263348556, or mute the thread https://github.com/notifications/unsubscribe-auth/AKKw-rAApYpdOmorvrbhjhIaGK03qRYPks5rCxougaJpZM4KeWVh .

[jessica-writes-code]

@all

Hey all -

I did a (reasonably brief) search for historical female employment statistics. Based on what I found, it seems that our best bet is something like this: http://www.bls.gov/cps/aa2006/aat9.txt, which is from the BLS. I think the idea would be to collect data for years where it's present and then map the BLS occupation categories to ours. That should be a reasonably straight-forward process (albeit a bit time-consuming and labor-intensive).

The major issues with this are:

• The occupations don't map especially well. E.g., "Life, physical, and social science occupations" probably includes some of the things we consider to be in "science & technology". However, it also probably includes plenty else that we don't want.
• There's a disjunction at 2001. Data don't seem to exist for the year 2000, AND the occupation categories change...
• It only goes back to 1995. That's fine, because there's 10 years of overlap, but obviously limits what conclusions we can draw from it.

Thoughts? Jessica