Skip to content

mubaris/urban-robot

Repository files navigation

Urban Robot

Urban Robot

Reddit bot which replies to sarcastic comments

Libraries

  • numpy, scipy - For Mathematical and Scientific processes
  • nltk - NLP Application
  • scikit - Model Training and Feature Extraction
  • textblob - Sentiment Analysis
  • pickle - Pickling Models and Vectorizers
  • langdetect - Language Detection of comments
  • praw - Reddit Bot

Features Used

  • Sentiment Analysis of full text, equal 2 and 3 parts of text
  • n-grams - 1 to 5
  • Term Frequency–Inverse Document Frequency(TF-IFD) after stemming, tokenizing and using n-grams of 1 to 5
  • Part of Speech Dictionary Vector
  • Topic Modeling

Data Preprocessing

  • Removed URLs
  • Removed Stopwords
  • Removed words with less than 4 tokens

Model Training and Classification

Using above Features and Preprocessing 4 models are trained,

  • Logistic Regression
  • Linear SVM
  • SVM with Gaussian Kernel
  • Random Forest

If a comment is predicted as 'sarcastic' by 3 out 4 models, it is treated as sarcastic.

Files

  • classifier.py - Training and Testing Models
  • bot.py - Reddit Bot
  • cli_bot.py - A Command Line Interactive Interface for the Reddit Bot
  • main.ipynb - iPython Notebook led to the final model hypothesis

Running

  1. Register for new Reddit App here and fill details (username, password, client id, client secret) under name 'bot1' in praw.ini

  2. Run classifier.py with Python 3(Optional) or use pretrained models

  3. Run bot.py with Python 3 for the automated Reddit Bot

  4. Run cli_bot.py with Python 3 for an interactive version of the Reddit Bot.

That's it.

Logs can accessed at comment.log

How to fill praw.ini

Final accuracy of models are in final_accuracy.txt

Dataset

Dataset is available in container

Downloaded from here