""" HuggingFace Space - ESS Variable Classification Demo Interactive Gradio interface for the XLM-RoBERTa ESS classifier. Developed by Sikt - Norwegian Agency for Shared Services in Education and Research """ import gradio as gr from transformers import pipeline # Load the model MODEL_NAME = "benjaminBeuster/xlm-roberta-base-ess-classification" classifier = pipeline("text-classification", model=MODEL_NAME) # Sikt brand colors SIKT_COLORS = { "amaranth": "#ee3243", # Primary accent "meteorite": "#331c6c", # Dark "selago": "#f3f1fe" # Light } # Category descriptions CATEGORY_INFO = { "DEMOGRAPHY (POPULATION, VITAL STATISTICS, AND CENSUSES)": "Demographics, population statistics, age, gender", "ECONOMICS": "Economic issues, finance, income, wealth", "EDUCATION": "Education, schooling, qualifications", "HEALTH": "Healthcare, medical services, health satisfaction", "POLITICS": "Political systems, trust in government, parliament", "SOCIETY AND CULTURE": "Social issues, cultural topics, religion", "LABOUR AND EMPLOYMENT": "Work, occupation, employment status", "PSYCHOLOGY": "Mental health, psychological wellbeing", "HOUSING AND LAND USE": "Housing conditions, residential environment", "NATURAL ENVIRONMENT": "Environmental concerns, climate change", "LAW, CRIME AND LEGAL SYSTEMS": "Justice, crime, legal matters", "MEDIA, COMMUNICATION AND LANGUAGE": "Media use, communication patterns", "SOCIAL STRATIFICATION AND GROUPINGS": "Social class, inequality, social groups", "SOCIAL WELFARE POLICY AND SYSTEMS": "Social benefits, welfare services", "TRANSPORT AND TRAVEL": "Transportation, mobility, travel patterns", "TRADE, INDUSTRY AND MARKETS": "Business, commerce, markets", "SCIENCE AND TECHNOLOGY": "Scientific advancement, technology use", "HISTORY": "Historical events, memory, heritage", "OTHER": "General or uncategorized topics" } def classify_text(text): """Classify survey question/variable.""" if not text.strip(): return "Please enter some text to classify." result = classifier(text)[0] label = result['label'] score = result['score'] # Format output output = f"**Category:** {label}\n\n" output += f"**Confidence:** {score:.2%}\n\n" if label in CATEGORY_INFO: output += f"**Description:** {CATEGORY_INFO[label]}" return output # Example questions - REAL questions from training data (metadata_classified.csv) # Randomly selected from actual ESS variables examples = [ ["How likely, governments in enough countries take action to reduce climate change"], ["Country"], ["Age of respondent, calculated"], ["Partner, control paid work last 7 days"], ["Ninth person in household: relationship to respondent"], ["Year of birth of eighth person in household"], ["Partner's age when completed full time education"], ["Religion or denomination belonging to in the past"], ["Which party feel closer to"], ["Highest level of education"], ["Ever unemployed and seeking work for a period more than three months"], ["Religion or denomination belonging to at present"], ["Partner doing last 7 days: housework, looking after children, others"], ["Year of birth of sixth person in household"], ["I like to be a leader, to what extent"], ["Main activity, last 7 days"], ["Mother's highest level of education"], ["Main activity last 7 days"], ["Doing last 7 days: unemployed, not actively looking for job"], ["How feminine respondent feels"], ["Father's highest level of education"], ["Trust in country's parliament"], ["How satisfied are you with the state of education"], ["How important to get respect from others"], ["Important to show abilities and be admired"], ["How often socially meet with friends, relatives or colleagues"], ["Placement on left right scale"], ["How often pray apart from at religious services"], ["Important to help people and care for others well-being"], ["Subjective general health"], ] # Custom CSS for Sikt branding using design tokens custom_css = """ :root { /* Sikt Design Tokens */ --sds-color-text-primary: #1a1a1a; --sds-color-text-secondary: #331c6c; --sds-color-interaction-primary: #7d5da6; --sds-color-interaction-primary-hover: #6b4d94; --sds-color-layout-background-default: #ffffff; --sds-color-layout-background-subtle: #f3f1fe; --sds-color-accent-primary: #ee3243; --sds-space-gap-small: 0.5rem; --sds-space-gap-medium: 1rem; --sds-space-gap-large: 1.5rem; --sds-space-padding-small: 0.75rem; --sds-space-padding-medium: 1rem; --sds-space-padding-large: 1.5rem; --sds-space-border-radius-small: 4px; --sds-space-border-radius-medium: 8px; --sds-space-border-radius-large: 12px; } .gradio-container { font-family: "Source Sans Pro", -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif !important; } h1, .gr-title { color: var(--sds-color-text-secondary) !important; font-weight: 600 !important; } .gr-box { border-radius: var(--sds-space-border-radius-medium) !important; } .gr-button { background-color: var(--sds-color-interaction-primary) !important; border-color: var(--sds-color-interaction-primary) !important; border-radius: var(--sds-space-border-radius-small) !important; font-weight: 500 !important; transition: all 0.2s ease !important; } .gr-button:hover { background-color: var(--sds-color-interaction-primary-hover) !important; border-color: var(--sds-color-interaction-primary-hover) !important; transform: translateY(-1px) !important; box-shadow: 0 2px 8px rgba(125, 93, 166, 0.3) !important; } .gr-button-primary { background: linear-gradient(135deg, var(--sds-color-interaction-primary) 0%, #6b4d94 100%) !important; } .gr-input, .gr-textbox { border-color: #e0e0e0 !important; border-radius: var(--sds-space-border-radius-small) !important; } .gr-input:focus, .gr-textbox:focus { border-color: var(--sds-color-interaction-primary) !important; box-shadow: 0 0 0 2px rgba(125, 93, 166, 0.1) !important; } .gr-panel { background-color: var(--sds-color-layout-background-subtle) !important; border-radius: var(--sds-space-border-radius-medium) !important; padding: var(--sds-space-padding-large) !important; } .gr-form { gap: var(--sds-space-gap-medium) !important; } footer { background-color: var(--sds-color-layout-background-subtle) !important; border-top: 1px solid #e0e0e0 !important; } .sikt-logo { max-width: 120px; height: auto; } .sikt-header { background: linear-gradient(135deg, #f3f1fe 0%, #ffffff 100%); padding: var(--sds-space-padding-medium); border-radius: var(--sds-space-border-radius-medium); margin-bottom: var(--sds-space-gap-large); border-left: 4px solid var(--sds-color-interaction-primary); } """ # Create Gradio interface with Sikt branding demo = gr.Interface( fn=classify_text, inputs=gr.Textbox( lines=3, placeholder="Enter a survey question or variable description...", label="Survey Question" ), outputs=gr.Markdown(label="Classification Result"), title="ESS Variable Classifier Prototype", description="""

ESS Variable Classifier Prototype

Developed by Sikt – Norwegian Agency for Shared Services in Education and Research

Automatically classify European Social Survey (ESS) questions into **19 subject categories**. This AI model is fine-tuned from XLM-RoBERTa-Base and achieves **83.8% accuracy**. **⚠️ Prototype Notice:** This model is trained on 582 samples. Only **8 categories** have reliable training data (≥20 samples): **Education, Politics, Society and Culture, Demography, Labour and Employment, Health, Psychology, and Other**. Results for other categories should be interpreted with caution. """, examples=examples, article="""

📊 About This Tool

This classifier helps researchers and data managers organize survey variables by automatically categorizing them into subject areas. The model was trained on European Social Survey metadata and can classify questions into 19 major categories from the CESSDA Topic Classification:

📚 EDUCATION 🏛️ POLITICS 🏥 HEALTH 💼 LABOUR AND EMPLOYMENT 🌍 SOCIETY AND CULTURE 💰 ECONOMICS 🧠 PSYCHOLOGY 👥 DEMOGRAPHY (POPULATION, VITAL STATISTICS, AND CENSUSES) 🏠 HOUSING AND LAND USE 🌱 NATURAL ENVIRONMENT ⚖️ LAW, CRIME AND LEGAL SYSTEMS 📺 MEDIA, COMMUNICATION AND LANGUAGE 👔 SOCIAL STRATIFICATION AND GROUPINGS 🤝 SOCIAL WELFARE POLICY AND SYSTEMS 🚗 TRANSPORT AND TRAVEL 🏪 TRADE, INDUSTRY AND MARKETS 🔬 SCIENCE AND TECHNOLOGY 📜 HISTORY 📋 OTHER

🔬 Technical Details

About Sikt

Sikt – Norwegian Agency for Shared Services in Education and Research provides digital infrastructure and services for research and education in Norway.

Visit sikt.no →

""", theme=gr.themes.Soft( primary_hue="red", secondary_hue="purple", ), css=custom_css ) if __name__ == "__main__": demo.launch()