Spaces:
Running
Running
| import gradio as gr | |
| import language_tool_python | |
| import nltk | |
| from transformers import pipeline | |
| from newspaper import Article | |
| from nltk.tokenize import sent_tokenize | |
| import re | |
| nltk.download("punkt") | |
| # Correct way to connect to the locally running LanguageTool server | |
| grammar_tool = language_tool_python.LanguageTool( | |
| language='en-US', | |
| remote_server='http://localhost:8081/v2/' | |
| ) | |
| # Hugging Face pipelines | |
| toxicity_classifier = pipeline("text-classification", model="unitary/toxic-bert") | |
| summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-6-6") | |
| def extract_text(input_type, text_input, url_input): | |
| if input_type == "URL" and url_input: | |
| article = Article(url_input) | |
| article.download() | |
| article.parse() | |
| return article.text | |
| return text_input | |
| def check_grammar(text): | |
| matches = grammar_tool.check(text) | |
| return [ | |
| { | |
| "text": match.context, | |
| "error": match.message, | |
| "suggestions": match.replacements, | |
| "offset": match.offset, | |
| "length": match.errorLength | |
| } for match in matches | |
| ] | |
| def detect_sensitive_content(text): | |
| sentences = sent_tokenize(text) | |
| sensitive_issues = [] | |
| for i, sentence in enumerate(sentences): | |
| result = toxicity_classifier(sentence) | |
| if result[0]['label'] == 'toxic' and result[0]['score'] > 0.7: | |
| sensitive_issues.append({ | |
| "sentence": sentence, | |
| "score": result[0]['score'], | |
| "index": i | |
| }) | |
| return sensitive_issues | |
| def generate_suggestions(text, grammar_issues, sensitive_issues): | |
| suggestions = [] | |
| for issue in grammar_issues: | |
| if issue['suggestions']: | |
| suggestions.append(f"Replace '{issue['text']}' with '{issue['suggestions'][0]}' ({issue['error']})") | |
| for issue in sensitive_issues: | |
| summary = summarizer(issue['sentence'], max_length=50, min_length=10, do_sample=False)[0]['summary_text'] | |
| suggestions.append(f"Rephrase sensitive content '{issue['sentence']}' to: '{summary}' (Toxicity score: {issue['score']:.2f})") | |
| return suggestions | |
| def highlight_text(text, grammar_issues, sensitive_issues): | |
| highlighted = text | |
| offset_adjust = 0 | |
| for issue in grammar_issues: | |
| start = issue['offset'] + offset_adjust | |
| end = start + issue['length'] | |
| error_text = highlighted[start:end] | |
| span = f"<span style='background-color: yellow'>{error_text}</span>" | |
| highlighted = highlighted[:start] + span + highlighted[end:] | |
| offset_adjust += len(span) - len(error_text) | |
| for issue in sensitive_issues: | |
| highlighted = highlighted.replace( | |
| issue['sentence'], | |
| f"<span style='background-color: red'>{issue['sentence']}</span>" | |
| ) | |
| return highlighted | |
| def review_blog(input_type, text_input, url_input): | |
| if not text_input and not url_input: | |
| return "Please provide text or a URL.", "", [] | |
| text = extract_text(input_type, text_input, url_input) | |
| grammar_issues = check_grammar(text) | |
| sensitive_issues = detect_sensitive_content(text) | |
| suggestions = generate_suggestions(text, grammar_issues, sensitive_issues) | |
| highlighted_text = highlight_text(text, grammar_issues, sensitive_issues) | |
| suggestions_text = "\n".join([f"{i+1}. {s}" for i, s in enumerate(suggestions)]) | |
| return highlighted_text, suggestions_text, suggestions | |
| def apply_changes(text, suggestions, approved_indices): | |
| sentences = sent_tokenize(text) | |
| for idx in approved_indices.split(','): | |
| try: | |
| idx = int(idx.strip()) - 1 | |
| if idx < len(suggestions): | |
| suggestion = suggestions[idx] | |
| match = re.search(r"'([^']+)'$", suggestion) | |
| if match: | |
| new_text = match.group(1) | |
| if "Rephrase sensitive content" in suggestion: | |
| orig_match = re.search(r"'([^']+)'\s+to:", suggestion) | |
| if orig_match: | |
| orig_sentence = orig_match.group(1) | |
| text = text.replace(orig_sentence, new_text) | |
| else: | |
| orig_match = re.search(r"Replace '([^']+)' with '([^']+)'", suggestion) | |
| if orig_match: | |
| text = text.replace(orig_match.group(1), orig_match.group(2)) | |
| except ValueError: | |
| continue | |
| return text | |
| with gr.Blocks(theme=gr.themes.Soft()) as demo: | |
| gr.Markdown("# ✨ AI Blog Reviewer") | |
| gr.Markdown("Highlight grammar & sensitive issues. Rephrase toxic content. Approve and apply changes.") | |
| 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) | |
| def toggle_input(type): | |
| return { | |
| text_input: gr.update(visible=type == "Text"), | |
| url_input: gr.update(visible=type == "URL") | |
| } | |
| input_type.change(fn=toggle_input, inputs=input_type, outputs=[text_input, url_input]) | |
| review_btn = gr.Button("Review Content") | |
| highlighted = gr.HTML() | |
| suggestions = gr.Textbox(label="Suggestions", lines=10) | |
| approved = gr.Textbox(label="Approve Suggestions (e.g., 1,3)") | |
| apply_btn = gr.Button("Apply Changes") | |
| final = gr.Textbox(label="Final Text", lines=10) | |
| suggestions_state = gr.State() | |
| review_btn.click(fn=review_blog, inputs=[input_type, text_input, url_input], | |
| outputs=[highlighted, suggestions, suggestions_state]) | |
| apply_btn.click(fn=apply_changes, inputs=[text_input, suggestions_state, approved], outputs=final) | |
| demo.launch() | |