Bot Detective Chrome Extension

Description

Bot Detective is a machine learning project that classifies Twitter users as bots or not bots based on user metadata. This project was developed to help Twitter users distinguish between real users and automated accounts, which can be used for spamming, spreading false information, and manipulating public opinion.

• Goals:
  • Raise bot awerness.
  • Enchanse user's security.
  • Promote a more cautious and critical approach when interacting with unfamiliar users.

Model Architecture

The user classification is performed by a Random Forest model, which utilizes the user's metadata collected through the Twitter API. The ML model we created is based on the article: "Scalable and Generalizable Social Bot Detection through Data Selection", it uses 4 diffrent classified datasets of Twitter users that were manually labeled as bots or not bots based on their metadata, such as the number of followers, following, tweets and likes. The model was developed using Python and various libraries such as pandas and scikit-learn. Dataset Building: The process of constructing the unified dataset is detailed in Build_Dataset.py. The resulting consolidated dataset is available at "Datasets/all_df.csv," while the individual datasets are stored within the "Datasets" folder. Model Construction: For the model creation procedure, you can refer to the Build_model.py. This script also encompasses an analysis of the model's performance against various alternative classifiers.

Features

Our extension includes 2 main features based on bot classification:

• Bot/Human indictor: As users browse through their Twitter feed, they will seamlessly come across visual indicators displaying "Bot" or "Human" sign—generated by the ongoing classification process of the model.
  

  

• Number of bots in 100 randomly chosen followers: For every user, an analysis of the amount of bots among 100 randomly selected followers is displayed in Profile page.
  We can see below an example of bot indicator and number of bots in 100 randomly chosen followers:

  

Paper

Technical Report

Requirements:

Python 3.9 with the following dependencies:

• fastapi
• uvicorn
• redis
• pickle
• pandas
• nltk.util
• requests
• numpy
• matplotlib
• sklearn.ensemble
• sklearn.model_selection
• sklearn.naive_bayes
• sklearn.calibration
• sklearn.inspection
• sklearn.metrics
• seaborn

Usage

For using the extension follow the next steps:

1. Clone reposetory.

  git clone https://github.com/stav-bentov/Bot-Detective.git

2. Open Chrome Extensions on developer mode.
3. Click on "Load unpacked" and choose the folder: extension in the cloned project.
4. To run our extension using the local server follow the next instructions:

• In API_requests.py: Make these lines in a comment:

app.add_middleware(HTTPSRedirectMiddleware)  # Redirect HTTP to HTTPS
...
uvicorn.run(app, host="0.0.0.0", port=3003, ssl_keyfile="./server.key", ssl_certfile="./server.crt")

Make sure this line is not in comment:

uvicorn.run(app, port=8000)

• In searchAndCalcUsers.js: Make these lines in a comment:

const response = await fetch(`https://34.165.1.66:3003/isBot/${Object.keys(usersOnRequestDict).join(',')}`);

Make sure this line is not in comment:

const response = await fetch(`http://127.0.0.1:8000/isBot/${Object.keys(usersOnRequestDict).join(',')}`);

• In showBotPrec.js: Make these lines in a comment:

response = await fetch(`https://34.165.1.66:3003/followersBots/?username=${username}&classification=${requestClassification}&followersPrec=${requestFollowersPrec}`);

Make sure this line is not in comment:

response = await fetch(`http://127.0.0.1:8000/followersBots/?username=${username}&classification=${requestClassification}&followersPrec=${requestFollowersPrec}`);

note: For running our Google Chrome server- do the opposite (regarding the comments) and make sure the server is activated.

5. Activate Redis in the server (local or our Google Cloud Server):

  sudo service redis-server start
  password: [enter password]
  redis-cli

6. Create a file in main directory named access_keys.py:

  consumer_key = __Fill you consumer key__
  consumer_secret = __Fill you consumer secret__
  bearer_token = __Fill you bearer token__
  access_token_key = __Fill you access token key__
  access_token_secret = __Fill you access token secret__

7. Run on the server:

    python API_requests.py

Or:

    python3 API_requests.py

8. Go back to Chrome Extension for developer and activate the extension.
9. You can browse through Twitter and have fun with Bot Detective.

Files/ Folders Descriptor

• /extension Code of extension, including manifest, content scripts and background script.
• API_requests.py The server, created by using Fast API, gets the requests and generates responses with the help of Detector.py that runs the model.
• Detector.py Uses the model and Twitter API for calculating results of classification and information about followers.
• Build_Dataset.py Building of united dataset: Datasets/all_df.csv, calculates metadata and dervied features.
• Build_Model.py Builds the Random Forest model based on the united dataset: Datasets/all_df.csv. This script includes grid search of Gaussian Naive Bayes, Random Forest and Linear SVC classifiers.
• /Datasets Includes classified Twitter users and united dataset all_df.csv.
• /Videos Videos of Bot Detective in run.
• /Articles,/Gifs,/Photos, /Knesset Folders that are not part of the technical develop of Bot Detctive, read README of them for more info.

Visual Overview

• User's classifications

While browsing- indicators appear, each accompanied by a popup displaying warnings and the classification accuracy (showed when mouse is over it).

• Informative Popup

When clicking the bot/human sign, an informative popup is displayed, providing details on scenarios where bots posed a risk by including quotes and links to news and articles.

• General Browsing Experience