Spaces:
Sleeping
Sleeping
| # imports | |
| import re | |
| import plotly.graph_objects as go | |
| import streamlit as st | |
| import json | |
| from openai import OpenAI | |
| import textstat | |
| import trafilatura | |
| import requests | |
| from bs4 import BeautifulSoup | |
| # constants | |
| MAX_WORDS = 400 | |
| ARTICLE_A = """In a watershed moment for global tech governance, international regulatory bodies have introduced the comprehensive Artificial Intelligence Safeguard Act. For too long, Silicon Valley titans have operated in a wild west environment, prioritizing unchecked corporate greed and rapid deployment over public safety. This landmark legislation aims to establish rigorous ethical boundaries and mandatory safety audits before any advanced generative models can be released to the public. Proponents argue that without these essential guardrails, society faces catastrophic risks ranging from massive, unmitigated job displacement to the proliferation of deepfake-fueled misinformation that threatens the very fabric of our democratic institutions. "We cannot allow a handful of unelected tech billionaires to play roulette with humanity's future," stated the coalition's lead ethicist. By prioritizing human welfare over blind technological acceleration, the Act serves as a vital moral firewall, ensuring that the development of artificial general intelligence benefits society as a whole rather than just enriching the elite few.""" | |
| ARTICLE_B = """Tech industry leaders and economists are sounding the alarm over the newly proposed Artificial Intelligence Safeguard Act, warning that the draconian legislation will severely cripple the nation’s economic engine. Critics argue that the bill is a masterclass in bureaucratic overreach, drowning agile tech startups in layers of punitive red tape and effectively stifling the very innovation that drives modern prosperity. By mandating arbitrary algorithmic audits and imposing heavy-handed restrictions on model training, the government is poised to surrender our global competitive edge to foreign adversaries who are not bound by such paralyzing regulations. "This isn't about safety; it's an innovation tax that penalizes success," argued a prominent venture capitalist. Analysts project that this short-sighted policy will force thousands of AI researchers to relocate overseas, draining billions of dollars in investment capital from the domestic market. Ultimately, framing technological progress as an inherent danger will only succeed in legislating the industry into obsolescence, destroying millions of future private-sector jobs in the process.""" | |
| URL_A = "https://www.foxnews.com/live-news/trump-iran-israel-war-updates-march-30" | |
| URL_B = "https://edition.cnn.com/2026/03/30/world/live-news/iran-war-us-israel-trump" | |
| st.set_page_config(page_title="FrameVis | Media Framing", layout="wide") | |
| # Initialize the AI model | |
| GROQ_API_KEY = st.secrets.get("GROQ_API_KEY") | |
| if not GROQ_API_KEY: | |
| st.warning("Groq API Token Missing.") | |
| st.stop() | |
| client = OpenAI( | |
| api_key=GROQ_API_KEY, | |
| base_url="https://api.groq.com/openai/v1", | |
| ) | |
| def _truncate_to_words(text: str, limit: int) -> str: | |
| """Truncates text by word count.""" | |
| if not text: | |
| return "" | |
| words = text.split() | |
| return " ".join(words[:limit]) | |
| def analyze_article(text: str) -> dict: | |
| """Analyzes framing using an LLM API and calculates readability.""" | |
| safe_text = _truncate_to_words(text, MAX_WORDS) | |
| prompt = f""" | |
| You are an expert media analyst. Analyze the following news snippet for framing, bias, and emotion. | |
| Return ONLY a valid JSON object with the exact keys below. Do not include markdown formatting or explanations. | |
| Keys to return: | |
| "sentiment_score": A float between -1.0 (highly negative) and 1.0 (highly positive). | |
| "primary_tone": The single dominant emotion. | |
| "primary_theme": Choose ONE like: ["economic consequences", "moral and ethical fairness", "legal and bureaucratic", "public safety and health"]. | |
| "tone_scores": A dictionary scoring THESE EXACT 6 EMOTIONS from 0.0 to 1.0: {{"anger": 0.0, "fear": 0.0, "joy": 0.0, "sadness": 0.0, "surprise": 0.0, "trust": 0.0}}. | |
| "framing_words": A list of dictionaries containing the 5 to 8 most emotionally charged or biased words, and the specific emotion they evoke. Format: [{{"word": "draconian", "emotion": "fear"}}, {{"word": "titans", "emotion": "awe"}}]. | |
| "subjectivity_score": A float between 0.0 (completely objective/factual) and 1.0 (highly opinionated/subjective). | |
| Text to analyze: | |
| "{safe_text}" | |
| """ | |
| try: | |
| response = client.chat.completions.create( | |
| model="llama-3.3-70b-versatile", | |
| response_format={"type": "json_object"}, | |
| temperature=0.1, | |
| messages=[ | |
| {"role": "system", "content": "You are a media analyst. You MUST respond with ONLY valid JSON matching the exact requested schema."}, | |
| {"role": "user", "content": prompt} | |
| ] | |
| ) | |
| llm_data = json.loads(response.choices[0].message.content) | |
| except Exception as e: | |
| print(f"Error parsing LLM response: {e}") | |
| llm_data = {} | |
| raw_reading_ease = textstat.flesch_reading_ease(safe_text) | |
| tones = llm_data.get("tone_scores", {}) | |
| standard_tones = { | |
| "anger": tones.get("anger", 0.0), | |
| "fear": tones.get("fear", 0.0), | |
| "joy": tones.get("joy", 0.0), | |
| "sadness": tones.get("sadness", 0.0), | |
| "surprise": tones.get("surprise", 0.0), | |
| "trust": tones.get("trust", 0.0), | |
| } | |
| return { | |
| "sentiment_score": llm_data.get("sentiment_score", 0.0), | |
| "primary_tone": llm_data.get("primary_tone", "neutral"), | |
| "primary_theme": llm_data.get("primary_theme", "unclear"), | |
| "tone_scores": standard_tones, | |
| "framing_words": llm_data.get("framing_words", []), | |
| "subjectivity_score": llm_data.get("subjectivity_score", 0.0), | |
| "reading_ease": max(0.0, min(100.0, raw_reading_ease)), | |
| } | |
| def _create_sentiment_gauge(score: float, title: str) -> go.Figure: | |
| """Generates a Plotly gauge chart for sentiment visualization.""" | |
| fig = go.Figure( | |
| go.Indicator( | |
| mode="gauge+number", | |
| value=score, | |
| domain={"x": [0, 1], "y": [0, 1]}, | |
| title={"text": title, "font": {"size": 16}}, | |
| gauge={ | |
| "axis": {"range": [-1, 1], "tickwidth": 1, "tickcolor": "darkgrey"}, | |
| "bar": {"color": "#475569", "thickness": 0.2}, | |
| "bgcolor": "white", | |
| "borderwidth": 0, | |
| "steps": [ | |
| {"range": [-1, -0.1], "color": "#fee2e2"}, | |
| {"range": [-0.1, 0.1], "color": "#f1f5f9"}, | |
| {"range": [0.1, 1], "color": "#dcfce3"}, | |
| ], | |
| }, | |
| ) | |
| ) | |
| fig.update_layout(height=280, margin=dict(l=20, r=20, t=60, b=20)) | |
| return fig | |
| def _create_comparison_radar_chart(results_a: dict, results_b: dict) -> go.Figure: | |
| """Generates an overlapping radar chart to compare emotions.""" | |
| categories = sorted(list(set(list(results_a["tone_scores"].keys()) + list(results_b["tone_scores"].keys())))) | |
| val_a = [results_a["tone_scores"].get(c, 0) for c in categories] | |
| val_b = [results_b["tone_scores"].get(c, 0) for c in categories] | |
| categories.append(categories[0]) | |
| val_a.append(val_a[0]) | |
| val_b.append(val_b[0]) | |
| fig = go.Figure() | |
| fig.add_trace(go.Scatterpolar( | |
| r=val_a, theta=categories, fill='toself', name='Source A', | |
| line=dict(color='#4f46e5', shape='spline', width=2), | |
| fillcolor='rgba(79, 70, 229, 0.2)' | |
| )) | |
| fig.add_trace(go.Scatterpolar( | |
| r=val_b, theta=categories, fill='toself', name='Source B', | |
| line=dict(color='#10b981', shape='spline', width=2), | |
| fillcolor='rgba(16, 185, 129, 0.2)' | |
| )) | |
| fig.update_layout( | |
| polar=dict( | |
| radialaxis=dict(visible=True, showticklabels=False, showline=False, gridcolor='rgba(0,0,0,0.1)'), | |
| angularaxis=dict(gridcolor='rgba(0,0,0,0.1)', linecolor='rgba(0,0,0,0.1)') | |
| ), | |
| showlegend=True, | |
| legend=dict(orientation="h", yanchor="bottom", y=-0.2, xanchor="center", x=0.5), | |
| title={"text": "Relative Emotion Profile", "font": {"size": 18, "family": "sans-serif"}}, | |
| height=400, | |
| margin=dict(l=40, r=40, t=60, b=40), | |
| paper_bgcolor='rgba(0,0,0,0)', # Transparent | |
| plot_bgcolor='rgba(0,0,0,0)' | |
| ) | |
| return fig | |
| def _highlight_framing_words(text: str, target_words: list) -> str: | |
| """Highlights LLM-identified framing words and tags their specific emotion.""" | |
| display_text = _truncate_to_words(text, MAX_WORDS) | |
| if not display_text: | |
| return "" | |
| highlighted_text = display_text + ("..." if len(text.split()) > MAX_WORDS else "") | |
| for item in target_words: | |
| if isinstance(item, dict): | |
| word = item.get("word", "") | |
| emotion = str(item.get("emotion", "charged")).upper() | |
| else: | |
| word = str(item) | |
| emotion = "CHARGED" | |
| if len(word) > 2: | |
| pattern = r'\b(' + re.escape(word) + r')\b' | |
| replacement = rf"""<span style='background-color: #f8fafc; border: 1px solid #cbd5e1; border-radius: 6px; padding: 0.15rem 0.3rem; font-weight: 600; color: #0f172a; box-shadow: 0 1px 2px rgba(0,0,0,0.02); margin: 0 0.1rem;'>\1 <span style='font-size: 0.65rem; font-weight: 700; text-transform: uppercase; letter-spacing: 0.05em; color: #ffffff; background-color: #475569; padding: 0.15rem 0.4rem; border-radius: 12px; margin-left: 4px; vertical-align: text-bottom;'>{emotion}</span></span>""" | |
| highlighted_text = re.sub(pattern, replacement, highlighted_text, flags=re.IGNORECASE) | |
| return highlighted_text | |
| def fetch_article_text(url: str) -> str: | |
| """Scrapes article text.""" | |
| downloaded = trafilatura.fetch_url(url) | |
| if downloaded: | |
| text = trafilatura.extract(downloaded) | |
| if text and len(text) > 200: | |
| return text | |
| try: | |
| headers = { | |
| 'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Safari/537.36', | |
| 'Accept-Language': 'en-US,en;q=0.9', | |
| } | |
| response = requests.get(url, headers=headers, timeout=10) | |
| soup = BeautifulSoup(response.content, 'html.parser') | |
| for script in soup(["script", "style"]): | |
| script.extract() | |
| paragraphs = soup.find_all('p') | |
| text = ' '.join([p.get_text() for p in paragraphs]) | |
| if text and len(text) > 200: | |
| return text.strip() | |
| except Exception as e: | |
| return f"Error: Could not fetch URL. Connection failed." | |
| return "Error: Could not extract text. The site may be protected by hard paywalls." | |
| def check_contradiction(text_a: str, text_b: str) -> dict: | |
| """Uses the LLM to evaluate the stance between arguments.""" | |
| safe_a = _truncate_to_words(text_a, MAX_WORDS) | |
| safe_b = _truncate_to_words(text_b, MAX_WORDS) | |
| prompt = f""" | |
| You are a fact-checking analyst. Compare these two news excerpts. | |
| Return ONLY a valid JSON object with the exact keys below. Do not include markdown formatting. | |
| Keys to return: | |
| "relationship": Choose ONE from: ["CONTRADICTION", "ENTAILMENT", "NEUTRAL"]. (Contradiction = disputing facts, Entailment = agreeing on premise). | |
| "confidence": A float between 0.0 and 1.0 representing how confident you are. | |
| Text 1: "{safe_a}" | |
| Text 2: "{safe_b}" | |
| """ | |
| try: | |
| response = client.chat.completions.create( | |
| model="llama-3.3-70b-versatile", | |
| response_format={"type": "json_object"}, | |
| temperature=0.1, | |
| messages=[ | |
| {"role": "system", "content": "You are a fact-checker. You MUST respond with ONLY valid JSON."}, | |
| {"role": "user", "content": prompt} | |
| ] | |
| ) | |
| result = json.loads(response.choices[0].message.content) | |
| except Exception as e: | |
| print(f"Error in contradiction check: {e}") | |
| result = {} | |
| return {"relationship": result.get("relationship", "NEUTRAL"), "confidence": result.get("confidence", 0.0)} | |
| def get_cached_analysis(text: str) -> dict: | |
| return analyze_article(text) | |
| def get_cached_contradiction(text_a: str, text_b: str) -> dict: | |
| return check_contradiction(text_a, text_b) | |
| st.markdown(""" | |
| <style> | |
| #MainMenu {visibility: hidden;} | |
| footer {visibility: hidden;} | |
| header {visibility: hidden;} | |
| .block-container { | |
| padding-top: 2rem; | |
| padding-bottom: 2rem; | |
| max-width: 1200px; | |
| } | |
| [data-testid="stMetric"] { | |
| background-color: #ffffff; | |
| border: 1px solid #e2e8f0; | |
| border-radius: 12px; | |
| padding: 20px; | |
| box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05), 0 2px 4px -1px rgba(0, 0, 0, 0.03); | |
| transition: transform 0.2s ease-in-out; | |
| } | |
| [data-testid="stMetric"]:hover { | |
| transform: translateY(-2px); | |
| box-shadow: 0 10px 15px -3px rgba(0, 0, 0, 0.1), 0 4px 6px -2px rgba(0, 0, 0, 0.05); | |
| } | |
| [data-testid="stMetricValue"], [data-testid="stMetricValue"] * { | |
| white-space: normal !important; | |
| overflow-wrap: break-word !important; | |
| word-wrap: break-word !important; | |
| overflow: visible !important; | |
| text-overflow: clip !important; | |
| line-height: 1.25 !important; | |
| } | |
| [data-testid="stMetricValue"] > div { | |
| color: #0f172a; | |
| font-weight: 700 !important; | |
| font-size: 1.3rem !important; | |
| } | |
| [data-testid="stMetricLabel"] > div { | |
| color: #64748b; | |
| font-weight: 500; | |
| text-transform: uppercase; | |
| letter-spacing: 0.05em; | |
| font-size: 0.85rem; | |
| } | |
| </style> | |
| """, unsafe_allow_html=True) | |
| # STATE MANAGEMENT | |
| if "results_a" not in st.session_state: | |
| st.session_state.results_a = None | |
| if "results_b" not in st.session_state: | |
| st.session_state.results_b = None | |
| if "nli_result" not in st.session_state: | |
| st.session_state.nli_result = None | |
| st.title("FrameVis") | |
| st.markdown("##### Media bias and framing effects across global news sources.") | |
| st.divider() | |
| with st.container(border=True): | |
| st.markdown("#### Configure Data Sources") | |
| input_method = st.radio("Input Method", ["Paste Text", "Paste URL"], horizontal=True, index=0) | |
| col1, col2 = st.columns(2) | |
| with col1: | |
| if input_method == "Paste Text": | |
| user_article_a = st.text_area("Data Source A", value=ARTICLE_A.strip(), height=180) | |
| else: | |
| url_a = st.text_input("Source A URL", value=URL_A) | |
| user_article_a = fetch_article_text(url_a) if url_a else "" | |
| with col2: | |
| if input_method == "Paste Text": | |
| user_article_b = st.text_area("Data Source B", value=ARTICLE_B.strip(), height=180) | |
| else: | |
| url_b = st.text_input("Source B URL", value=URL_B) | |
| user_article_b = fetch_article_text(url_b) if url_b else "" | |
| st.write("") | |
| execute_analysis = st.button("Analyze and Compare Sources", use_container_width=True, type="primary") | |
| if execute_analysis: | |
| text_a_clean = user_article_a.strip() if user_article_a else "" | |
| text_b_clean = user_article_b.strip() if user_article_b else "" | |
| if not text_a_clean or not text_b_clean: | |
| st.warning("Please provide text or a valid URL for both Source A and Source B before analyzing.") | |
| elif text_a_clean.startswith("Error:") or text_b_clean.startswith("Error:"): | |
| st.error("One of the URLs could not be scraped. Please copy and paste the text directly.") | |
| else: | |
| with st.spinner("Analyzing both sources."): | |
| try: | |
| res_a = get_cached_analysis(text_a_clean) | |
| res_b = get_cached_analysis(text_b_clean) | |
| res_nli = get_cached_contradiction(text_a_clean, text_b_clean) | |
| st.session_state.results_a = res_a | |
| st.session_state.results_b = res_b | |
| st.session_state.nli_result = res_nli | |
| except Exception as e: | |
| error_msg = str(e) | |
| if "429" in error_msg or "RESOURCE_EXHAUSTED" in error_msg: | |
| st.error("**Rate Limit Reached:** Too many requests too quickly.") | |
| else: | |
| st.error(f"API or Processing Error: {error_msg}") | |
| st.session_state.results_a = None | |
| st.session_state.results_b = None | |
| st.session_state.nli_result = None | |
| # Analysis Display | |
| if st.session_state.results_a and st.session_state.results_b: | |
| st.divider() | |
| st.markdown("### Framing Analytics & Comparison") | |
| # Display Contradictions | |
| nli_result = st.session_state.nli_result | |
| if nli_result: | |
| if nli_result["relationship"].upper() == "CONTRADICTION": | |
| st.error(f"**NARRATIVE CONTRADICTION** (Confidence: {nli_result['confidence']:.2f}) - These sources are disputing each other's facts.") | |
| elif nli_result["relationship"].upper() == "ENTAILMENT": | |
| st.success(f"**NARRATIVE ALIGNMENT** (Confidence: {nli_result['confidence']:.2f}) - These sources agree on the core premise.") | |
| else: | |
| st.info(f"**NEUTRAL RELATIONSHIP** - These sources are discussing the topic without direct contradiction or alignment.") | |
| st.plotly_chart(_create_comparison_radar_chart(st.session_state.results_a, st.session_state.results_b), use_container_width=True) | |
| res_col1, res_col2 = st.columns(2) | |
| # Render Column A | |
| with res_col1: | |
| r_a = st.session_state.results_a | |
| st.markdown("#### Source A Breakdown") | |
| m1, m2 = st.columns(2) | |
| m3, m4 = st.columns(2) | |
| m1.metric("Subjectivity", f"{r_a['subjectivity_score']:.2f}", help="0 is objective, 1 is highly opinionated.") | |
| m2.metric("Primary Emotion", str(r_a['primary_tone']).title()) | |
| m3.metric("Framing Lens", str(r_a['primary_theme']).title()) | |
| m4.metric("Reading Ease", f"{r_a['reading_ease']:.1f}", help="0-30 is college graduate level, 60-70 is 8th grade.") | |
| st.plotly_chart(_create_sentiment_gauge(r_a["sentiment_score"], "Sentiment Bias"), use_container_width=True, key="gauge_a") | |
| st.markdown("**Key Framing Language:**") | |
| annotated_text = _highlight_framing_words(user_article_a, r_a['framing_words']) | |
| st.markdown(f"<div style='background-color: #ffffff; padding: 1.5rem; border-radius: 12px; border: 1px solid #e2e8f0; line-height: 1.8; font-size: 1.05rem; box-shadow: 0 1px 3px rgba(0,0,0,0.05);'>{annotated_text}</div>", unsafe_allow_html=True) | |
| # Render Column B | |
| with res_col2: | |
| r_b = st.session_state.results_b | |
| st.markdown("#### Source B Breakdown") | |
| m1, m2 = st.columns(2) | |
| m3, m4 = st.columns(2) | |
| m1.metric("Subjectivity", f"{r_b['subjectivity_score']:.2f}", help="0 is objective, 1 is highly opinionated.") | |
| m2.metric("Primary Emotion", str(r_b['primary_tone']).title()) | |
| m3.metric("Framing Lens", str(r_b['primary_theme']).title()) | |
| m4.metric("Reading Ease", f"{r_b['reading_ease']:.1f}", help="0-30 is college graduate level, 60-70 is 8th grade.") | |
| st.plotly_chart(_create_sentiment_gauge(r_b["sentiment_score"], "Sentiment Bias"), use_container_width=True, key="gauge_b") | |
| st.markdown("**Key Framing Language:**") | |
| annotated_text = _highlight_framing_words(user_article_b, r_b['framing_words']) | |
| st.markdown(f"<div style='background-color: #ffffff; padding: 1.5rem; border-radius: 12px; border: 1px solid #e2e8f0; line-height: 1.8; font-size: 1.05rem; box-shadow: 0 1px 3px rgba(0,0,0,0.05);'>{annotated_text}</div>", unsafe_allow_html=True) |