Uses NLP techniques to hunt for potential bot/shill/sock puppet accounts on Twitter.
The application is currently using the Vader package to generate sentiment analysis values for input search terms. Also returned is an average (mean) compound value for each search term. This is displayed on the screen to the user, but will in the future be output as a file for futher statistical analysis if desired.
A config file takes the following format:
config:
access_token: '<twitter access token>'
access_secret: '<twitter access secret>
consumer_key: '<twitter consumer key>'
consumer_secret: '<twitter consumer secret>'
target: <twitter username>
search_terms:
- Hurricane
- Irma
dialect: en-US
The first four values can be found in your Twitter developer account:
The target value should be a twitter username or ID.
Following this is a list of search terms listed under the
search_terms key.
The final value is dialect e,g, en-US
Based upon the dialect input, tweets will be searched to see if
they deviate from target dialect.
For example if the target account is suppose to be using UK English but uses US spellings and grammer, this will be flagged.
Dialects are stored within the package and UK and US English are currently supported, with a small list of words currently being matched.
Spelling errors are flagged which is very noisy as URLs and other strings will be listed. This option may be useful if looking for patterns in spelling mistakes or punctuation errors across a series of different twitter accounts.
Clone the source code from git and install via pip.
pip install -e python-twitter-shill-hunter
Then run
python -m twitter_shill_hunter twitter.yaml <list of plugins>
Here the plugins can be: sentiment_analysis, grammar_analysis, spelling_analysis
For example:
python -m twitter_shill_hunter twitter.yaml sentiment_analysis