search

barik / sharelatex-ihearthn

1 stars 1 forks source link

Automatically measure bias #8

Closed barik closed 9 years ago

barik commented 9 years ago

many social news / media websites contain rich information about their users, both explicit (e.g, demographics information, but also self-assessment of expertise, interest, etc) and implicit (e.g., based on Q&A one can derive expertise, engagement, etc). i wonder if one could automatically measure bias in the sampled data using such information, so to avoid having to do :random" sampling as done in section 4 (where 100 comments were randomly selected out of 600)

CaptainEmerson commented 9 years ago

I'm pretty sure I used the word "blowhard" to describe some comments in early comments on this draft. :)

barik commented 9 years ago

I think this comment is actually right, but my inclination right now is that it's out of scope for the NIER paper. It's definitely something I wanted to look at in the journal paper, especially metadata signals such as user karma as a proxy for trust. I might add a sentence about this during the sampling sentence.

CaptainEmerson commented 9 years ago

Ok. I wonder if there are nlp techniques for detecting hyperbole. On Jul 14, 2015 1:02 AM, "Titus Barik" notifications@github.com wrote:

I think this comment is actually right, but my inclination right now is that it's out of scope for the NIER paper. It's definitely something I wanted to look at in the journal paper, especially metadata signals such as user karma as a proxy for trust. I might add a sentence about this during the sampling sentence.

— Reply to this email directly or view it on GitHub https://github.com/barik/sharelatex-ihearthn/issues/8#issuecomment-121133562 .

barik commented 9 years ago

The space is not going to permit this discussion, which at this point seems orthogonal to the main idea anyway.