How biased is the news org/ person?

This image was the inspiration for this project. I wanted to make the X axis of this:
Like this:

My idea:

1. Define a right and left wing corpus with words that those wings are usually associated with
   1. Manually decide 20-30 accounts (politicians, commentators, spokespersons etc..) each from right and left wing for loud polarised people
   2. Use Twitter API to input tweets from them
   3. For every right-wing tweet:
      1. Run GCP entity sentiment analysis
      2. Add entities with positive sentiment to right wing corpus
      3. Add entities with negative sentiment to left wing corpus
      4. There will be entities that are in both corpuses (‘army’, ‘mahatma gandhi’ etc).
      5. Do vice-versa for left-wing tweets
2. Now stream news like I did in my other project
3. Run GCP entity sentiment analysis on news data
4. Compare these entity sentiments with the above defined corpora
5. Calculate difference in % match
6. Scatter plot and clustering of different news sources

These corpora would be at a national level, not same corpora for all international sources

Oversimplifying the scenario into a BJP vs Congress world. These are the twitter accounts I have used to build the corpus.
## INC
### List of INC spokespersons: [Official](https://https://twitter.com/i/lists/200025713/members)
['JM_Scindia', 'RajBabbarMP', '_SandeepDikshit', 'SATAVRAJEEV', 'VijayIndrSingla', 'Meem_Afzal', 'MYaskhi', 'SinghRPN', 'sushmitadevinc', 'dineshgrao', 'plpunia', 'GauravGogoiAsm', 'DrAMSinghvi', 'drajoykumar', 'khushsundar', 'shaktisinhgohil', 'JhaSanjay', 'DeependerSHooda', 'rajeevgowda']
### Others
['RahulGandhi', 'sunilkjakhar', 'srinivasiyc', 'INCIndiaLive']
---
## BJP
### List of BJP spokespersons: (Verified accounts)
['sambitswaraj', 'ShahnawazBJP', 'NupurSharmaBJP', 'gauravbh', 'GVLNRAO', 'AshwiniBJP', 'NalinSKohli']
### Others
['BJP4India', 'rammadhavbjp', 'JPNadda', 'BJPLive']

Did not include cabinet ministers since their tweets are overwhelmingly about their portfolios and not about the common narrative.

This could be extended to do the same for transcripts from YouTube videos of NDTV, Republic TV etc (a data science heavy project) It could also be used to validate apparent bias in platforms, Eg - FB (right) and Twitter (left), and also gauge how various subsections (read subreddits) are aligned on our scale.

Relevant reads:

1. https://www.cjr.org/innovations/measure-media-bias-partisan.php - Good intro read for the concept. Specifically see the ‘armchair academics’ section
2. Similar manual projects:
   1. https://www.adfontesmedia.com/interactive-media-bias-chart/ for interactive link to source of 2D image above
   2. https://www.journalism.org/2014/10/21/political-polarization-media-habits/pj_14-10-21_mediapolarization-08/ link to source of 1D image above
   3. https://www.allsides.com/media-bias/media-bias-ratings
   4. Can use these to validate our outputs

Unrelated, but interesting - http://graphics.wsj.com/blue-feed-red-feed/#/president-trump

Progress so far

I have completed upto Step 1.3.3 (described above)

Here is a small sample of what I extracted for tweets from a congress party spokesperson :

After failing to contain the large scale migration,the BJP Govt is now indulging in inhuman act by forcing labourers to take bath in chemical solution.

This is shocking & deplorable that the UP Police & BJP Govt is using such tactics in the name of preventing corona outbreak. https://t.co/l2Iodi2Qcd

— Rajeev Satav (@SATAVRAJEEV) March 30, 2020

You can find the notebook to do this over here. I’m retiring this project due to time constraints but if you’d like to complete this project, the main steps remaining are #4, #5 and #6, data science related.

Check out my other projects!🌟