SMM-CompromisedAccountDetection

Setup

Python packages

You have to install the required packages using pip:

pip install -r requirements.txt

NLTK Data

The application additionally needs the following NLTK data packages:

• punkt - Punkt Tokenizer Models

For installation instructions please visit http://www.nltk.org/data.html.

Usage

Command Line Interface

python cli.py [options]

Options

General arguments

The first option is the switch between the crawl cli and the analyze cli:

• crawl crawls tweets from the (default: 100) most popular twitter users and stores them on disk.
• analyze analyzes social media status updates in order to determine, whether an account was compromised or not.
• evaluate evaluates the anomaly detection approach with cross-validation.

Crawling status updates

• -o /--output-path OUTPUT_PATH The output path of the generated dataset.
• --user-limit USER_LIMIT The maximum number of accounts to crawl.
• --limit LIMIT The maximum number of status updates per user to crawl. (default: 100)

Analyzing status updates

• -ut / --user-data-source USER_DATA_SOURCE The data source for tweets of the user that should be analyzed. Possible values are fth, mp and twitter.
• -uu / --user-twitter-id USER_TWITTER_ID The id of the twitter user, whose status updates should be analyzed.
• -up / --user-dataset-path USER_DATASET_PATH The path of the dataset of the user data source.
• -et / --ext-data-source EXT_DATA_SOURCE The data source for external tweets not written by the user. Possible values are fth, mp and twitter.
• -ep / --ext-dataset-path EXT_DATASET_PATH The path of the dataset of the external data source.
• -c / --classifier-type CLASSIFIER_TYPE The type of the classifier to be trained. Possible values are decision_tree, one_class_svm, isolation_forest and perceptron.

Evaluation

• -t / --data-source DATA_SOURCE The data source for tweets that should be used for cross-validation. Possible values are fth, mp and twitter.

• -p / --dataset-path DATASET_PATH The path of the dataset that should be used for cross-validation.

• -c / --classifier-type CLASSIFIER_TYPE The type of the classifier to be trained. Possible values are decision_tree, one_class_svm, isolation_forest and perceptron.

  parser.add_argument("--dataset-path", "-p", help="The path of the dataset of the user data source. ") parser.add_argument("--classifier-type", "-c", help="The type of the classifier to be trained. ")

Examples

# Crawl 50 most popular users
python cli.py -a crawl -o output.csv --user-limit 50

# Analyze twitter account of sebsatian_kliem against status updates from the 'Follow the Hashtag' dataset using the perceptron classifier
python cli.py analyze -ut twitter -uu sebastian_kliem -et fth -ep data/follow_the_hashtag_usa.csv -c perceptron

Web App

# Run the app locally (DO NOT use this in production)
./run_app_dev.sh

The app takes a twitter url from a specific user as input and uses the perceptron classifier.