f

app.py

@@ -0,0 +1,88 @@
+import streamlit as st
+import numpy as np
+import joblib
+import re
+from keras.models import load_model
+from keras.preprocessing.sequence import pad_sequences
+
+# Define max_sequence_length (same as what was used during training)
+max_sequence_length = 186
+
+# Define bad_words list
+bad_words = ["মাদারচুদ", "বাইন চুদ", "জুকার", "ফালতু", "শালা", "লেংটা", "বাটপার", "মাদারচুদ", "ফালতু", "শালা", "নাস্তিকের বাচ্চা", "শুয়ার", "কুত্তা", "পুটকি", "নগ্নতায়", "সমকামি", "চুদছে", "চুদতে", "চুদা", "আবাল চোদা", "শুয়োরের বাচ্চা", "কুত্তার বাচ্চা", "হারামির বাচ্চা", "হারামজাদা", "শালার পো", "চুতমারানি", "চুদির ভাই","হাউয়ার নাতি", "খানকি", "মারা", "হোগা", "খানকির পোলা", "চোদা", "মিয়া খলিফা", "জনি সিন্স", "মাগির পোলা", "মাগি", "মাগী", "পর্ণ", "গরুচোদা", "হিজরার", "হিজরা"]
+
+# Load the model
+loaded_model = load_model('cyberbullying_model.h5')
+
+# Load the tokenizer and label encoder
+loaded_tokenizer = joblib.load('tokenizer.pkl')
+loaded_label_encoder = joblib.load('label_encoder.pkl')
+
+# Function to filter out bad words from text and return filtered words
+def filter_bad_words_with_model(text):
+    filtered_words = []
+    for bad_word in bad_words:
+        # Create a regular expression pattern for the bad word, ignoring case
+        pattern = re.compile(re.escape(bad_word), re.IGNORECASE)
+        # Replace occurrences of the bad word with asterisks (*) of the same length
+        text, num_replacements = pattern.subn('*' * len(bad_word), text)
+        if num_replacements > 0:
+            filtered_words.append(bad_word)
+    return text, filtered_words
+
+
+# Streamlit UI
+st.title("Cyberbullying Detection App (Bangla)")
+
+# Input text for prediction
+input_text = st.text_area("Enter Text:")
+
+if st.button("Predict"):
+    if input_text:
+        # Filter out bad words using the loaded model
+        filtered_text, filtered_bad_words = filter_bad_words_with_model(input_text)
+
+        # Tokenize and pad the filtered input text
+        input_sequence = loaded_tokenizer.texts_to_sequences([filtered_text])
+        input_sequence = pad_sequences(input_sequence, maxlen=max_sequence_length)
+
+        # Make a prediction using the loaded model
+        predicted_probabilities = loaded_model.predict(input_sequence)
+
+        # Get the class with the highest probability as the predicted class
+        predicted_class = np.argmax(predicted_probabilities)
+
+        # Decode the predicted class back to the original label using the loaded label encoder
+        predicted_label = loaded_label_encoder.classes_[predicted_class]
+
+        st.subheader("Prediction Result:")
+        if predicted_label == "not bully":
+            st.write("Prediction: Not Cyberbullying")
+            st.write("No bad words found.")
+        else:
+            st.write("Prediction: Cyberbullying")
+            st.write(f"Cyberbullying Type: {predicted_label}")
+
+        if filtered_bad_words: 
+            st.write(f"Bad Words: {', '.join(filtered_bad_words)}")
+        else:
+            st.write("<span style='color:cyan;'>No bad words found.</span>", unsafe_allow_html=True)
+
+        if filtered_bad_words:
+            st.write("Filtered Text:")
+            filtered_text_with_asterisks = filtered_text
+            for bad_word in filtered_bad_words:
+                filtered_text_with_asterisks = re.sub(re.escape(bad_word), '*' * len(bad_word), filtered_text_with_asterisks, flags=re.IGNORECASE)
+
+            st.write(f"<span style='color:red; font-weight:bold'>{filtered_text_with_asterisks}</span>", unsafe_allow_html=True)
+        else:
+            st.write("Original Text:")
+            st.write(f"{input_text}", unsafe_allow_html=True)
+
+
+
+
+
+st.header("Sample Texts")
+st.write("ভেবেছিলাম তুই একটা ছেলে!!! এখন দেখি এটা একটা" + "<span style='color:red; font-weight:bold'> হিজরা</span>?", unsafe_allow_html=True)
+st.write("প্রতিটি নাটক কয়েকবার করে দেখা হয়ে গেছে")

cyberbullying_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afc432e360c18e5cf63c4dee6cf0d3120fddf09189baefa229b7968c5c98d6be
+size 8645176

label_encoder.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9eed52d22388d26d10e2032d8aaf6210f839b604f77fd039a8e2f6220e8eabc
+size 582

requirements.txt

@@ -0,0 +1,8 @@
+scikit-learn
+joblib
+regex
+keras
+tensorflow
+numpy
+streamlit
+dill

tokenizer.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a77dc749ed802387c54a13ff7af56dc210594b1b9b03797da5de60615048a733
+size 3424620