Comment-Toxicity-Analysis

Index

• Dataset
• Initial Stage
  • About Files
  • Fault
• Final Engine
• How to Configure Final Engine
• Improvement Scope
• Helpful Refrenceses

Dataset

• DataSet Download Site
• Engine Notebook Kaggle

Initial Stage

• This Initial Model is just for Comparing or for starting Without Preprocessing and Ensembling techniques.
• It tells toxicity after Analysing the Comment.
  • If a comment is Toxic or Bad then it will return Toxic else it will return toxic.

About Files

• ToxityAnalysis.ipynb -> For building model
• finalized_model.pkl -> Saved ML model
• RunModel.py -> For Reloading model and analysing Comment
• app.js -> For Running nodejs app

Fault

• It take more time because it reload python script evertime on pressing Submitting the Comment.

Final Engine

• Here is the engine of the model which is hosted using Flask.

• It Resolved the Fault of Initial Stage Model.

• It Only takes time on first time loading.

• It Returns the Percentage instead of Class.

• you Can See the Jupyter-notebook https://github.com/ckshitij/Comment-Toxicity-Analysis/blob/master/engine/Models/notebook.ipynb

  • Here its gives the Toxicity Percentage of the Comment.

  

  • You Can see the Difference between the Percentage of the comment

  

How to Configure Final Engine

• First Download the code directly from here in zip or clone through git.

• Then go to the engine folder.

• Click On the above Engine Notebook Kaggle and download all the .pkl file in Models folder.

• Then to to the Terminal and run app.py file.

• Then open your Web Browser and type http://localhost:5000/.

  git clone https://github.com/ckshitij/Comment-Toxicity-Analysis.git
  cd Comment-Toxicity-Analysis
  cd engine
  python3 app.py
  

Improvement Scope

• You Can improve this Code by Increasing the n-gram range (1,3or4) for Word-Level and n-gram range (1,10) for Char-Level.
• You can use Deeplearning RNN (LSTM).
• You can use word level Cnn Model too.

Helpful Refrenceses

• google Perspective
• kaggle Competition on Toxicity
• githu mediawiki-utilities
• Detecting Insults in Social Commentary Kaggle
• Content Analysis
• Abusive Language Detection in Online User Content
• Beware of Publicity! Perceived Distress of Negative Cyber Incidents and Implications for Defining Cyberbullying
• Learning part-of-speech taggers with inter-annotator agreement loss