ambanum / DormantTwitterAccountsDetector

[HACKATHON] Detect "dormant" bots on Twitter

Detect dormant bots on Twitter within the network of some given account.

Features of a dormant account

Keep in mind that this is just a hackathon proof of concept. No one is saying that any account bearing these features is necessarily a dormant bot. However, manual classification has proven that even with such naive conditions, around 85% of detected accounts do indeed seem to be inauthentic.

• Max 15 tweets
• Ratio Friends/Followers > 3
• Created after Feb 01 2019

Configuration

In the file .env:

• Specify your Twitter credentials:

    CONSUMER_KEY=<TWITTER_CONSUMER_KEY>
    CONSUMER_SECRET=<TWITTER_CONSUMER_SECRET>
    ACCESS_TOKEN=<TWITTER_ACCESS_TOKEN>
    ACCESS_TOKEN_SECRET=<TWITTER_ACCESS_TOKEN_SECRET>

• Specify the MongoDB url:

    MONGO_URL=mongodb://localhost/TwitterGraphExplorer

• Specify the Twitter handle of the seed user, that is the user from whom the graph exploration will start:

    USER_SEED=ndpnt

Usage

Install dependencies:

    npm install

Insert the seed user:

    node firstUserToExplore.js

Start the detector:

    node jobsRunner.js

You also can start a web UI to display a counter of dormant bots found:

    npm start

Result

As result, the script stores every account matching the given conditions in a twitterusers collection. It also stores every encountered account in a backup collection to allow changing features and rerun the detector locally without calling the Twitter API.