Topic {topic_id+1}
"""
# Sort words by weight for better visualization
word_weight_pairs = sorted(zip(words, weights), key=lambda x: x[1], reverse=True)
# Add each word with size based on weight
for word, weight in word_weight_pairs:
# Scale weight to a reasonable font size (min 14px, max 28px)
font_size = 14 + min(14, round(weight * 30))
# Color based on weight (darker = higher weight)
color_intensity = max(0, min(90, int(100 - weight * 100)))
color = f"hsl(210, 70%, {color_intensity}%)"
word_cloud_html += f"""
{word}
"""
word_cloud_html += """
"""
output_components.append(gr.HTML(word_cloud_html))
# Add a proper bar chart visualization for topic words
for topic in topics[:min(3, len(topics))]: # Show charts for max 3 topics to avoid clutter
topic_id = topic.get("id", 0)
words = topic.get("words", [])
weights = topic.get("weights", [])
if words and weights and len(words) == len(weights):
# Create dataframe for plotting
df = pd.DataFrame({
'word': words,
'weight': weights
})
# Sort by weight
df = df.sort_values('weight', ascending=False)
# Limit to top N words for clarity
df = df.head(10)
# Create bar chart
fig = px.bar(
df, x='weight', y='word',
title=f"Topic {topic_id+1} Top Words",
labels={'word': 'Word', 'weight': 'Weight'},
height=300,
orientation='h' # Horizontal bars
)
# Improve layout
fig.update_layout(
margin=dict(l=10, r=10, t=40, b=10),
yaxis={'categoryorder': 'total ascending'}
)
output_components.append(gr.Plot(value=fig))
# Visualize topic distributions for each model
model_topics = topic_results.get("model_topics", {})
if model_topics and all(model in model_topics for model in models):
output_components.append(gr.Markdown("### Topic Distribution by Model"))
# Create multi-model topic distribution comparison
distribution_data = []
for model in models:
if model in model_topics:
distribution = model_topics[model]
for i, weight in enumerate(distribution):
if i < 10: # Limit to 10 topics max
distribution_data.append({
'Model': model,
'Topic': f"Topic {i+1}",
'Weight': weight
})
if distribution_data:
df = pd.DataFrame(distribution_data)
# Create grouped bar chart
fig = px.bar(
df, x='Topic', y='Weight', color='Model',
barmode='group',
title="Topic Distribution Comparison",
height=400
)
output_components.append(gr.Plot(value=fig))
# Visualize topic differences as a heatmap
comparisons = topic_results.get("comparisons", {})
if comparisons:
comparison_key = f"{models[0]} vs {models[1]}"
if comparison_key in comparisons:
output_components.append(gr.Markdown("### Topic Similarity Analysis"))
# Get JS divergence
js_divergence = comparisons[comparison_key].get("js_divergence", 0)
# Create a divergence meter
divergence_html = f"""
Topic Distribution Divergence
Similar (0.0)
Different (1.0)
Score: {js_divergence:.3f}
Jensen-Shannon Divergence measures the similarity between topic distributions.
Lower values indicate more similar topic distributions between models.
"""
output_components.append(gr.HTML(divergence_html))
# Create similarity matrix heatmap if available
similarity_matrix = topic_results.get("similarity_matrix", [])
if similarity_matrix and len(similarity_matrix) > 0:
# Convert to format for heatmap
z_data = similarity_matrix
# Create heatmap
fig = go.Figure(data=go.Heatmap(
z=z_data,
x=[f"{models[1]} Topic {i+1}" for i in range(len(similarity_matrix[0]))],
y=[f"{models[0]} Topic {i+1}" for i in range(len(similarity_matrix))],
colorscale='Viridis',
showscale=True,
colorbar=dict(title="Similarity")
))
fig.update_layout(
title="Topic Similarity Matrix",
height=400,
margin=dict(l=50, r=50, t=50, b=50)
)
output_components.append(gr.Plot(value=fig))
# Show best matching topics
matched_topics = topic_results.get("matched_topics", [])
if matched_topics:
output_components.append(gr.Markdown("### Most Similar Topic Pairs"))
# Create HTML table for matched topics
matched_topics_html = """