Editing HCDS (Fall 2017)/Assignments (section)

=== A2: Bias in data ===

The goal of this assignment is to explore the concept of 'bias' through data on Wikipedia articles - specifically, articles on political figures from a variety of countries. For this assignment, you will combine a dataset of Wikipedia articles with a dataset of country populations, and use a machine learning service called ORES to estimate the quality of each article.

You are expected to perform an analysis of how the ''coverage'' of politicians on Wikipedia and the ''quality'' of articles about politicians varies between countries. Your analysis will consist of a series of tables that show:
# the countries with the greatest and least coverage of politicians on Wikipedia compared to their population. 
# the countries with the highest and lowest proportion of high quality articles about politicians.

You are also expected to write a short reflection on the project, that describes how this assignment helps you understand the causes and consequences of bias on Wikipedia.

==== Getting the article and population data ====

The first step is getting the data, which lives in several different places. The wikipedia dataset can be found [https://figshare.com/articles/Untitled_Item/5513449 on Figshare]. Read through the documentation for this repository, then download and unzip it. 

The population data is on the [http://www.prb.org/DataFinder/Topic/Rankings.aspx?ind=14 Population Research Bureau website]. Download this data as a CSV file (hint: look for the 'Microsoft Excel' icon in the upper right). 

==== Getting article quality predictions ====

Now you need to get the predicted quality scores for each article in the Wikipedia dataset. For this step, we're using a Wikimedia API endpoint for a machine learning system called [https://www.mediawiki.org/wiki/ORES ORES] ("Objective Revision Evaluation Service"). ORES estimates the quality of an article (at a particular point in time), and assigns a series of probabilities that the article is in one of 6 quality categories. The options are, from best to worst:

# FA - Featured article
# GA - Good article
# B - B-class article
# C - C-class article
# Start - Start-class article
# Stub - Stub-class article

For context, these quality classes are a sub-set of quality assessment categories developed by Wikipedia editors. If you're curious, you can read more about what these assessment classes mean on [https://en.wikipedia.org/wiki/Wikipedia:WikiProject_assessment#Grades English Wikipedia]. We will talk about what these categories mean, and how the ORES model predicts which category an article goes into, next week in class. For this assignment, you only need to know that these categories exist, and that ORES will assign one of these 6 categories to any article you send it.

The ORES API is configured fairly similarly to the pageviews API we used last assignment; documentation can be found [https://ores.wikimedia.org/v3/#!/scoring/get_v3_scores_context_revid_model here]. It expects a revision ID, which is the third column in the Wikipedia dataset, and a model, which is "wp10". The sample iPython notebook for this assignment provides an example of a correctly-structured API query that you can use to understand how to gather your data, and also to examine the query output.

In order to get article predictions for each article in the Wikipedia dataset, you will need to read <tt>page_data.csv</tt> into Python (or R), and then read through the dataset line by line, using the value of the <tt>last_edit</tt> column in the API query. If you're working in Python, the [https://docs.python.org/3/library/csv.html CSV module] will help with this.

When you query the API, you will notice that ORES returns a <tt>prediction</tt> value that contains the name of one category, as well as <tt>probability</tt> values for each of the 6 quality categories. For this assignment, you only need to capture and use the value for <tt>prediction</tt>. We'll talk more about what the other values mean in class next week.

==== Combining the datasets ====
 
Some processing of the data will be necessary! In particular, you'll need to - after retrieving and including the ORES data for each article - merge the wikipedia data and population data together. Both have fields containing country names for just that purpose. After merging the data, you'll invariably run into entries which ''cannot'' be merged. Either the population dataset does not have an entry for the equivalent Wikipedia country, or vice versa. You will need to remove the rows that do not have matching data.

Consolidate the remaining data into a single CSV file which looks something like this:


{|class="wikitable"
|-
! Column
|-
|country
|-
|article_name
|-
|revision_id
|-
|article_quality
|-
|population
|}

Note: <tt>revision_id</tt> here is the same thing as <tt>last_edit</tt>, which you used to get scores from the ORES API.

==== Analysis ====
Your analysis will consist of calculating the proportion (as a percentage) of articles-per-population and high-quality articles for each country. By "high quality" articles, in this case we mean the number of articles about politicians in a given country that ORES predicted would be in either the "FA" (featured article) or "GA" (good article) classes.

Examples:
* if a country has a population of 10,000 people, and you found 10 articles about politicians from that country, then the percentage of articles-per-population would be .1%. 
* if a country has 10 articles about politicians, and 2 of them are FA or GA class articles, then the percentage of high-quality articles would be 20%. 

==== Tables ====
The tables should be pretty straightforward. Produce four tables that show:
#10 highest-ranked countries in terms of number of politician articles as a proportion of country population
#10 lowest-ranked countries in terms of number of politician articles as a proportion of country population
#10 highest-ranked countries in terms of number of GA and FA-quality articles as a proportion of all articles about politicians from that country
#10 lowest-ranked countries in terms of number of GA and FA-quality articles as a proportion of all articles about politicians from that country

Embed them in the iPython notebook.

==== Writeup ====
Write a few paragraphs, either in the README or in the notebook, reflecting on what you have learned, what you found, what (if anything) surprised you about your findings, and/or what theories you have about why any biases might exist (if you find they exist). You can also include any questions this assignment raised for you about bias, Wikipedia, or machine learning.

==== Submission instructions ====
#Complete your Notebook and datasets in Jupyter Hub.
#Create the data-512-a2 repository on GitHub w/ your code and data.
#Complete and add your README and LICENSE file.
#Submit the link to your GitHub repo to: https://canvas.uw.edu/courses/1174178/assignments/3876068

==== Required deliverables ====
A directory in your GitHub repository called <tt>data-512-a2</tt> that contains the following files:
:# 1 final data file in CSV format that follows the formatting conventions.
:# 1 Jupyter notebook named <tt>hcds-a2-bias</tt> that contains all code as well as information necessary to understand each programming step, as well as your writeup (if you have not included it in the README) and the tables.
:# 1 README file in .txt or .md format that contains information to reproduce the analysis, including data descriptions, attributions and provenance information, and descriptions of all relevant resources and documentation (inside and outside the repo) and hyperlinks to those resources, and your writeup (if you have not included it in the notebook).
:# 1 LICENSE file that contains an [https://opensource.org/licenses/MIT MIT LICENSE] for your code.

==== Helpful tips ====
* Read all instructions carefully before you begin
* Read all API documentation carefully before you begin
* Experiment with queries in the sandbox of the technical documentation for the API to familiarize yourself with the schema and the data
* Explore the data a bit before starting to be sure you understand how it is structured and what it contains
* Ask questions on Slack if you're unsure about anything
* When documenting/describing your project, think: "If I found this GitHub repo, and wanted to fully reproduce the analysis, what information would I want? What information would I need?"