ishaan007/woah_emnlp_2020

Code for hate filtering heuristics from twitter archive

A Novel Methodology for Developing Automatic Harassment Classifiers for Twitter

Research conducted by Columbia University Speech Lab

Part of the Fourth Workshop on Online Abuse and Harms (co-hosted with EMNLP)

Development done by:

• Ishaan Arora (MS '20, Columbia CS)
• Julia Guo (BA '22, Columbia CS)

Advised by:

• Sarah Ita Levitan (Professor, Hunter College; Previous postdoctorate, Columbia CS)
• Susan E. McGregor (Professor, Columbia Journalism & CS)
• Julia Hirschberg (Professor, Columbia CS)

Overview

In this repository, we have included core code components used to fetch tweet threads potentially containing hate speech from Twitter archive data. The primary heuristics currently used to fetch this data are:

• Get threads containing blocked and muted users from the target's Twitter archive
• Get threads containing subtweets (mention of real name, but not of username) from the Twitter Search API

Structure

driver.py: Main driver script for scraping data from Twitter archive and Twitter Search API

utils.py: Helper Tweet retrieval functions used in driver.py and ways_to_fetch_tweet_threads.py

ways_to_fetch_tweet_threads.py: Classes for fetching tweet threads (from Archive, Search, ...)

Requirements

• tweepy = 3.9.0

Setting up environment

conda create -n <env_name>
conda activate <env_name>
pip install -r requirements.txt

How to run code with Twitter archive data

First, modify the function get_auth_cred() in utils.py with your appropriate Twitter Developer API keys. If you don't already have these, or are unfamiliar with the Twitter API, please see the below links.

• Twitter Developer API application
• Tweepy documentation

Next, learn how to download your Twitter Archive here.

After everything is set up and ready to go, you can run driver.py using Twitter archive files (tweet, mute, block) in the following manner:

python driver.py --tweet_file=<tweet file path> --mute_file=<mute file path> --block_file=<block file path> --real_name=<real name> --user_name=<twitter username>

You can also provide optional arguments that specify the number of tweets to fetch using each method (muted, blocked, non-muted/non-blocked, subtweets). The default values are 100, 100, 200, 100, respectively.

Here is an example command to run:

python driver.py \
  --tweet_file="jd_tweet.js" \
  --mute_file="jd_mute.js" \
  --block_file="jd_block.js" \
  --real_name="Jane Doe" \
  --user_name="JaneDoe" \
  --mute_tweets_ct=10 \
  --block_tweets_ct=20 \
  --other_tweets_ct=30 \
  --subtweet_tweets_ct=40

After the script finishes running, it will create a new data file, dump/filtered.json, containing tweet threads pulled from both the provided Twitter archive files, and the Search API.

To add a new tweet filtering heuristic:

Define a new subclass of TweetThreadsFromSource in ways_to_fetch_tweet_threads.py

Example:

class TweetThreadsFromSearch(TweetThreadsFromSource):

Override the method get_tweet_threads_list in this newly defined class.

Contact information:

Ishaan Arora: [email protected]
Julia Guo: [email protected]

woah_emnlp_2020's Issues

Enhancement: driver.py file should run with default arguments for number of tweets in different files

If user does not provide number of tweets for mute.js, blocked.js and tweet.js. The script should run on entire file. Current implementation forces the user to find out total number of tweets in each file, if they want to run it on entire file

Handle edge case in driver.py - take uniques

Before final call to serialize_tweets() is made in driver.py. There is a bleak chance that tweet threads from multiple sources get replicated. In that case a small enhancement to find uniques of list of list of dictionaries needs to be made.