Chatbots that help to fight disinformation on Mattermost.

Integrations

BotometerAnalyser

Assesses the distribution of robots-like accounts sharing specific content on Twitter. It uses Botometer but instead of analysing one account, it analyses all accounts that have shared specific content.

This integration is called through Mattermost as a Slash command with a search term and returns a distribution of the “botscores”, i.e. the how robot-like the Twitter accounts that have tweeted the search term behave. As we only use the Twitter Standard search API, the search is made against a sampling of recent Tweets published in the past 7 days with a mix between recent and popular content.

Example result in Mattermost where a slash command is set up to be used as /botometer [search term]:

Installation

After cloning the repository, install cairo:

On ubuntu:

    sudo apt update
    sudo apt install libcairo2-dev libjpeg-dev libgif-dev redis

Then:

    npm install

See Troubleshooting in case of errors

Configuration

Configure each hook in the proper environment file, for example in production.json:

{
    "hooks": {
        "domain": "https://desinfo.quaidorsay.fr/bots/",
        "botometerAnalyser": {
            "twitter": {
                "consumer_key": "<CONSUMER_KEY>",
                "consumer_secret": "<CONSUMER_SECRET>",
                "access_token": "<ACCESS_TOKEN>",
                "access_token_secret": "<ACCESS_TOKEN_SECRET>"
            },
            "botometer": {
                "consumer_key": "<CONSUMER_KEY>",
                "consumer_secret": "<CONSUMER_SECRET>",
                "access_token_key": "<ACCESS_TOKEN_KEY>",
                "access_token_secret": "<ACCESS_TOKEN_SECRET>",
                "mashape_key": "<MASHAPE_KEY>"
            }
        }
    }
}

hooks.domain: Base URL used to make distribution graphs available.

Usage

In order to start a server that responds to incoming requests from a Mattermost instance:

    npm start

Dev

Create a file .env:

TWITTER_CONSUMER_KEY=xxx
TWITTER_CONSUMER_SECRET=xxx
TWITTER_ACCESS_TOKEN_KEY=xxx
TWITTER_ACCESS_TOKEN_SECRET=xxx
MASHAPE_KEY=xxx

Install docker. Run a mattermost image: docker run --name mattermost-preview -d --publish 8065:8065 --add-host dockerhost:127.0.0.1 mattermost/mattermost-preview

Access the local mattermost instance at localhost:8065. Create a new team, a new channel, an integration (slash command) to http://host.docker.internal:3000/botometer (GET). In "system console"/"advanced"/"developer", add the trusted hosts 127.0.0.1 localhost host.docker.internal.

Install redis and run a server: redis-server.

Create the file config/development.js:

{
    "hooks": {
        "domain": "http://localhost:3000",
        "botometerAnalyser": {
            "mattermost": {
                "token": "xxx"
            }
        },
        "sendToAnalysis": {
            "incomingWebHookUrl": "http://localhost:8065/hooks/xxx",
            "actionUrl": "http://host.docker.internal:3000/media-scale",
            "actionResponseUrl": "http://localhost:8065/hooks/xxx"
        }
    }
}

To run the automated tests:

npm test

To run manual tests:

DEBUG=BotometerAnalyser:* npm start

Linux

The magic hostname host.docker.internal does not work on linux systems yet.

Run docker run --name mattermost-preview -d --publish 8065:8065 --network="host" mattermost/mattermost-preview, access at address http://127.0.0.1:8065, set the integration to http://127.0.0.1:3000/botometer.

Troubleshoohting

Failed at the canvas@1.6.13 install script.

xcode-select --install
brew install pkg-config cairo pango libpng jpeg giflib librsvg

License

EUPL v1.2: akin to an AGPL, but more readable and translated and legally binding into all languages of the EU. Recap.