Update app.py

app.py

@@ -5,111 +5,83 @@ import nltk
 from nltk.tokenize import sent_tokenize
 import re
 
-# Download punkt tokenizer
-nltk.download("punkt")
+# Download punkt tokenizer for sentence splitting
+nltk.download('punkt')
 
-# Load Hugging Face pipelines
-grammar_corrector = pipeline("text2text-generation", model="pszemraj/flan-t5-base-grammar-synthesis")
-toxicity_detector = pipeline("text-classification", model="unitary/toxic-bert")
+# Load grammar correction and toxicity detection models
+grammar_corrector = pipeline("text2text-generation", model="vennify/t5-base-grammar-correction")
+toxicity_classifier = pipeline("text-classification", model="unitary/toxic-bert")
 
-# Utility: extract blog text from URL
 def extract_text(input_type, text_input, url_input):
-    if input_type == "Text":
-        return text_input
-    elif input_type == "URL":
+    if input_type == "URL" and url_input:
         try:
             article = Article(url_input)
             article.download()
             article.parse()
             return article.text
         except Exception as e:
-            return f"Error extracting article: {str(e)}"
-    return ""
-
-# Highlight grammar & sensitive content
-def highlight_text(text, grammar_sentences, toxic_sentences):
-    for s in grammar_sentences:
-        text = text.replace(s, f"<span style='background-color: yellow'>{s}</span>")
-    for s in toxic_sentences:
-        text = text.replace(s, f"<span style='background-color: red'>{s}</span>")
-    return text
-
-# Main blog review function
-def review_blog(input_type, text_input, url_input):
-    text = extract_text(input_type, text_input, url_input)
-    if not text or text.startswith("Error"):
-        return text, "", []
+            return f"Error extracting from URL: {str(e)}"
+    return text_input
+
+def check_grammar(text):
+    try:
+        corrected = grammar_corrector(text, max_length=512)[0]['generated_text']
+        return corrected
+    except Exception as e:
+        return f"Error in grammar correction: {str(e)}"
 
+def detect_sensitive_content(text):
     sentences = sent_tokenize(text)
-    grammar_issues = []
-    toxic_issues = []
-    suggestions = []
-
-    for sent in sentences:
-        # Check grammar by comparing original and corrected
-        corrected = grammar_corrector(sent, max_length=128, do_sample=False)[0]['generated_text']
-        if corrected.strip() != sent.strip():
-            grammar_issues.append(sent)
-            suggestions.append(f"Grammar: Replace '{sent}' → '{corrected}'")
-
-        # Check for toxicity
-        result = toxicity_detector(sent)
+    sensitive = []
+    for i, sentence in enumerate(sentences):
+        result = toxicity_classifier(sentence)
         if result[0]['label'] == 'toxic' and result[0]['score'] > 0.7:
-            toxic_issues.append(sent)
-            suggestions.append(f"Toxicity: Rephrase '{sent}' (score: {result[0]['score']:.2f})")
+            sensitive.append({
+                "sentence": sentence,
+                "score": result[0]['score'],
+                "index": i
+            })
+    return sensitive
+
+def highlight_sensitive(text, sensitive_issues):
+    highlighted = text
+    for issue in sensitive_issues:
+        sentence = issue['sentence']
+        highlighted = highlighted.replace(sentence, f"<span style='background-color:red'>{sentence}</span>")
+    return highlighted
 
-    highlighted = highlight_text(text, grammar_issues, toxic_issues)
-    sug_text = "\n".join(f"{i+1}. {s}" for i, s in enumerate(suggestions))
-    return highlighted, sug_text, suggestions
+def review_blog(input_type, text_input, url_input):
+    text = extract_text(input_type, text_input, url_input)
+    if text.startswith("Error"):
+        return text, text
 
-# Apply approved suggestions
-def apply_changes(original_text, suggestions, indices):
-    try:
-        indices = [int(i.strip()) - 1 for i in indices.split(",") if i.strip().isdigit()]
-        sentences = sent_tokenize(original_text)
-        for i in indices:
-            if i < len(suggestions):
-                match = re.search(r"'(.*?)'\s*→\s*'(.*?)'", suggestions[i])
-                if match:
-                    old = match.group(1)
-                    new = match.group(2)
-                    original_text = original_text.replace(old, new)
-        return original_text
-    except Exception as e:
-        return f"Error applying changes: {str(e)}"
+    corrected_text = check_grammar(text)
+    sensitive_issues = detect_sensitive_content(corrected_text)
+    highlighted = highlight_sensitive(corrected_text, sensitive_issues)
+
+    return highlighted, corrected_text
 
 # Gradio UI
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("## ✨ Blog Content Reviewer (LLM-powered)")
-    gr.Markdown("Paste blog content or a blog URL. AI will detect grammar issues & sensitive content.")
+    gr.Markdown("# Blog Grammar & Toxicity Reviewer")
 
-    input_type = gr.Radio(["Text", "URL"], value="Text", label="Input Type")
+    input_type = gr.Radio(["Text", "URL"], label="Input Type", value="Text")
     text_input = gr.Textbox(label="Blog Text", lines=10, visible=True)
     url_input = gr.Textbox(label="Blog URL", visible=False)
 
-    input_type.change(lambda t: {text_input: gr.update(visible=t=="Text"),
-                                 url_input: gr.update(visible=t=="URL")},
-                      input_type, [text_input, url_input])
-
-    review_button = gr.Button("Review")
-    highlighted_output = gr.HTML()
-    suggestions_output = gr.Textbox(label="Suggestions", lines=10)
-    approve_input = gr.Textbox(label="Approve Suggestions (e.g., 1,2)")
-    apply_button = gr.Button("Apply Changes")
-    final_text = gr.Textbox(label="Final Output", lines=10)
+    def toggle_inputs(choice):
+        return {
+            text_input: gr.update(visible=choice == "Text"),
+            url_input: gr.update(visible=choice == "URL")
+        }
 
-    suggestions_state = gr.State()
+    input_type.change(fn=toggle_inputs, inputs=input_type, outputs=[text_input, url_input])
 
-    review_button.click(
-        fn=review_blog,
-        inputs=[input_type, text_input, url_input],
-        outputs=[highlighted_output, suggestions_output, suggestions_state]
-    )
+    review_button = gr.Button("Review Blog")
+    highlighted_output = gr.HTML(label="Highlighted Output (Red = Sensitive)")
+    corrected_output = gr.Textbox(label="Corrected Text", lines=10)
 
-    apply_button.click(
-        fn=apply_changes,
-        inputs=[text_input, suggestions_state, approve_input],
-        outputs=[final_text]
-    )
+    review_button.click(fn=review_blog, inputs=[input_type, text_input, url_input], outputs=[highlighted_output, corrected_output])
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()