app.py

import streamlit as st
import pandas as pd
import numpy as np
import pickle
from PIL import Image
# preprocessing
import re
import string
import nltk
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer 
from sklearn.feature_extraction.text import CountVectorizer
# modeling
from sklearn import svm
# sentiment analysis
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer

# creating page sections
site_header = st.container()
business_context = st.container()
data_desc = st.container()
performance = st.container()
tweet_input = st.container()
model_results = st.container()
sentiment_analysis = st.container()
contact = st.container()

with site_header:
    st.title('Twitter Hate Speech Detection')
   
    
    This project aims to **automate content moderation** to identify hate speech using **machine learning binary classification algorithms.** 
    
    Baseline models included Random Forest, Naive Bayes, Logistic Regression and Support Vector Machine (SVM). The final model was a **Logistic Regression** model that used Count Vectorization for feature engineering. It produced an F1 of 0.3958 and Recall (TPR) of 0.624.  

with business_context:
    st.header('The Problem of Content Moderation')
    st.write("""
    
    **Human content moderation exploits people by consistently traumatizing and underpaying them.** In 2019, an [article](https://www.theverge.com/2019/6/19/18681845/facebook-moderator-interviews-video-trauma-ptsd-cognizant-tampa) on The Verge exposed the extensive list of horrific working conditions that employees faced at Cognizant, which was Facebook’s primary moderation contractor. Unfortunately, **every major tech company**, including **Twitter**, uses human moderators to some extent, both domestically and overseas.
    
    Hate speech is defined as **abusive or threatening speech that expresses prejudice against a particular group, especially on the basis of race, religion or sexual orientation.**  Usually, the difference between hate speech and offensive language comes down to subtle context or diction.
    
    """)

with data_desc:
    understanding, venn = st.columns(2)
    with understanding:
        st.text('')
        st.write("""
        The **data** for this project was sourced from a Cornell University [study](https://github.com/t-davidson/hate-speech-and-offensive-language) titled *Automated Hate Speech Detection and the Problem of Offensive Language*.
        
        The `.csv` file has **24,802 rows** where **6% of the tweets were labeled as "Hate Speech".**
        Each tweet's label was voted on by crowdsource and determined by majority rules.
        """)
        
with tweet_input:
    st.header('Is Your Tweet Considered Hate Speech?')
    st.write("""*Please note that this prediction is based on how the model was trained, so it may not be an accurate representation.*""")
    # user input here
    user_text = st.text_input('Enter Tweet', max_chars=280) # setting input as user_text

with model_results:    
    st.subheader('Prediction:')
    if user_text:
    # processing user_text
        # removing punctuation
        user_text = re.sub('[%s]' % re.escape(string.punctuation), '', user_text)
        # tokenizing
        stop_words = set(stopwords.words('english'))
        tokens = nltk.word_tokenize(user_text)
        # removing stop words
        stopwords_removed = [token.lower() for token in tokens if token.lower() not in stop_words]
        # taking root word
        lemmatizer = WordNetLemmatizer() 
        lemmatized_output = []
        for word in stopwords_removed:
            lemmatized_output.append(lemmatizer.lemmatize(word))

        # instantiating count vectorizor
        count = CountVectorizer(stop_words=stop_words)
        X_train = pickle.load(open('X_train', 'rb'))
        X_test = lemmatized_output
        X_train_count = count.fit_transform(X_train)
        X_test_count = count.transform(X_test)

        # loading in model
        final_model = pickle.load(open('bayes', 'rb'))

        # apply model to make predictions
        prediction = final_model.predict(X_test_count[0])

        if prediction == 0:
            st.subheader('**Not Hate Speech**')
        else:
            st.subheader('**Hate Speech**')
        st.text('')