Spaces:
Sleeping
Sleeping
File size: 22,219 Bytes
4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 82cacb5 4da2489 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 |
import gradio as gr
import numpy as np
import plotly.graph_objects as go
import plotly.express as px
import pandas as pd
import math
import random
from datetime import datetime, timedelta
import json
# ---------------------------
# GLOBAL DATA (SIMULATION)
# ---------------------------
TOPICS = ["AI/Tech", "Music", "Travel", "Food", "Gaming", "Fitness", "Fashion", "Art", "Business", "Comedy"]
# More complex global trends with temporal variations
GLOBAL_TRENDING = {
"AI/Tech": {"base": 0.95, "volatility": 0.1, "seasonal": 1.2},
"Music": {"base": 0.90, "volatility": 0.15, "seasonal": 1.1},
"Travel": {"base": 0.88, "volatility": 0.2, "seasonal": 0.8},
"Food": {"base": 0.85, "volatility": 0.05, "seasonal": 1.0},
"Gaming": {"base": 0.82, "volatility": 0.12, "seasonal": 1.0},
"Fitness": {"base": 0.80, "volatility": 0.08, "seasonal": 1.3},
"Fashion": {"base": 0.87, "volatility": 0.18, "seasonal": 1.1},
"Art": {"base": 0.75, "volatility": 0.10, "seasonal": 0.9},
"Business": {"base": 0.78, "volatility": 0.07, "seasonal": 0.95},
"Comedy": {"base": 0.92, "volatility": 0.20, "seasonal": 1.0}
}
# Content creator tiers
CREATOR_TIERS = {
"Micro": {"followers": 1000, "engagement_mult": 1.2, "reach_mult": 0.8},
"Mid": {"followers": 50000, "engagement_mult": 1.0, "reach_mult": 1.0},
"Macro": {"followers": 500000, "engagement_mult": 0.8, "reach_mult": 1.5},
"Celebrity": {"followers": 5000000, "engagement_mult": 0.6, "reach_mult": 2.0}
}
# User demographics
DEMOGRAPHICS = ["Gen Z", "Millennial", "Gen X", "Boomer"]
REGIONS = ["North America", "Europe", "Asia", "South America", "Africa"]
np.random.seed(42)
# More complex embedding space (higher dimensional)
EMBEDDING_DIM = 8
POST_EMBEDDINGS = {}
USER_DEMOGRAPHICS = {}
for topic in TOPICS:
base_trend = GLOBAL_TRENDING[topic]["base"]
volatility = GLOBAL_TRENDING[topic]["volatility"]
embedding = np.random.randn(EMBEDDING_DIM) * base_trend + np.random.randn(EMBEDDING_DIM) * volatility
POST_EMBEDDINGS[topic] = embedding
GLOBAL_EMBEDDING = np.mean(list(POST_EMBEDDINGS.values()), axis=0)
# ---------------------------
# ENHANCED UTILS
# ---------------------------
def sigmoid(x, steepness=1.0):
return 1 / (1 + math.exp(-steepness * x))
def attention_mechanism(query, keys, values, temperature=1.0):
"""Simplified attention mechanism for content ranking"""
scores = np.dot(keys, query) / temperature
weights = np.exp(scores) / np.sum(np.exp(scores))
return np.dot(weights, values), weights
def temporal_decay(time_diff_hours, half_life=24):
"""Content freshness decay"""
return 0.5 ** (time_diff_hours / half_life)
def diversity_penalty(selected_topics, candidate_topic, penalty_strength=0.1):
"""Penalty for showing too much of the same content"""
count = selected_topics.count(candidate_topic)
return math.exp(-penalty_strength * count)
# ---------------------------
# ENHANCED VALUE MODEL
# ---------------------------
def enhanced_value_model(like_signal, comment_signal, share_signal, save_signal,
watch_time, creator_tier, recency_hours, user_history_match):
"""More sophisticated value model with multiple signals"""
# Base probabilities
p_like = sigmoid(like_signal - 1, steepness=0.8)
p_comment = sigmoid(comment_signal - 0.5, steepness=1.2)
p_share = sigmoid(share_signal - 0.3, steepness=1.5)
p_save = sigmoid(save_signal - 0.2, steepness=1.0)
p_watch = sigmoid(watch_time - 2, steepness=0.6)
# Creator influence
creator_mult = CREATOR_TIERS[creator_tier]["engagement_mult"]
# Recency factor
recency_factor = temporal_decay(recency_hours)
# User history matching
history_boost = sigmoid(user_history_match - 0.5, steepness=2.0)
# Weighted scoring
base_score = (
1.0 * p_like +
3.0 * p_comment +
4.0 * p_share +
2.5 * p_save +
1.5 * p_watch
)
# Apply modifiers
final_score = base_score * creator_mult * recency_factor * (0.5 + 0.5 * history_boost)
return round(final_score, 4), {
"P(Like)": round(p_like, 3),
"P(Comment)": round(p_comment, 3),
"P(Share)": round(p_share, 3),
"P(Save)": round(p_save, 3),
"P(Watch)": round(p_watch, 3),
"Creator Multiplier": round(creator_mult, 3),
"Recency Factor": round(recency_factor, 3),
"History Match": round(history_boost, 3)
}
# ---------------------------
# ENHANCED COLD START WITH EXPLORATION
# ---------------------------
def enhanced_cold_start(interactions, preferred_topics, demographics, region,
exploration_factor=0.2):
"""Enhanced cold start with demographic targeting and exploration"""
# Blending factor based on interactions
alpha = min(interactions / 50.0, 0.9) # More gradual transition
# Create personal embedding from multiple preferences
if len(preferred_topics) > 0:
personal_embeddings = [POST_EMBEDDINGS[topic] for topic in preferred_topics]
personal_embedding = np.mean(personal_embeddings, axis=0)
else:
personal_embedding = GLOBAL_EMBEDDING
# Demographic influence
demo_noise = np.random.randn(EMBEDDING_DIM) * 0.1
if demographics == "Gen Z":
demo_noise += np.array([0.2, -0.1, 0.3, 0.1, 0.2, -0.1, 0.1, 0.2])
elif demographics == "Millennial":
demo_noise += np.array([0.1, 0.2, 0.1, 0.2, 0.1, 0.2, 0.1, 0.0])
# Regional trends
regional_bias = np.zeros(EMBEDDING_DIM)
if region == "Asia":
regional_bias += np.array([0.3, 0.1, -0.2, 0.2, 0.1, 0.0, 0.1, 0.1])
elif region == "North America":
regional_bias += np.array([0.1, 0.2, 0.1, 0.1, 0.3, 0.1, 0.2, 0.1])
# Exploration component (random discovery)
exploration_noise = np.random.randn(EMBEDDING_DIM) * exploration_factor
# Final blended embedding
user_embedding = (
alpha * personal_embedding +
(1 - alpha) * GLOBAL_EMBEDDING +
demo_noise +
regional_bias +
exploration_noise
)
return user_embedding
# ---------------------------
# CONTENT RANKING SYSTEM
# ---------------------------
def rank_content_feed(user_embedding, content_pool_size=20, diversity_weight=0.3):
"""Simulate full feed ranking with diversity considerations"""
# Generate synthetic content
content_items = []
for i in range(content_pool_size):
topic = random.choice(TOPICS)
creator_tier = random.choices(
list(CREATOR_TIERS.keys()),
weights=[40, 35, 20, 5]
)[0]
# Content embedding with some noise
content_emb = POST_EMBEDDINGS[topic] + np.random.randn(EMBEDDING_DIM) * 0.1
# Relevance score (cosine similarity)
relevance = np.dot(user_embedding, content_emb) / (
np.linalg.norm(user_embedding) * np.linalg.norm(content_emb)
)
# Random engagement signals
engagement_signals = {
"likes": max(0, np.random.normal(5, 2)),
"comments": max(0, np.random.normal(2, 1)),
"shares": max(0, np.random.normal(1, 0.5)),
"saves": max(0, np.random.normal(0.8, 0.3)),
"watch_time": max(0, np.random.normal(4, 1.5)),
"recency": np.random.uniform(0.1, 48)
}
# Calculate value score
value_score, _ = enhanced_value_model(
engagement_signals["likes"],
engagement_signals["comments"],
engagement_signals["shares"],
engagement_signals["saves"],
engagement_signals["watch_time"],
creator_tier,
engagement_signals["recency"],
max(0, relevance)
)
content_items.append({
"id": i,
"topic": topic,
"creator_tier": creator_tier,
"relevance": relevance,
"value_score": value_score,
"embedding": content_emb,
**engagement_signals
})
# Rank with diversity
ranked_items = []
remaining_items = content_items.copy()
selected_topics = []
for position in range(min(10, len(remaining_items))):
best_item = None
best_score = -float('inf')
for item in remaining_items:
# Combined score: relevance + diversity
diversity_score = diversity_penalty(selected_topics, item["topic"])
combined_score = (
(1 - diversity_weight) * item["value_score"] +
diversity_weight * diversity_score
)
if combined_score > best_score:
best_score = combined_score
best_item = item
if best_item:
ranked_items.append(best_item)
selected_topics.append(best_item["topic"])
remaining_items.remove(best_item)
return ranked_items
# ---------------------------
# ENHANCED UI FUNCTIONS
# ---------------------------
def tab_enhanced_value_model(likes, comments, shares, saves, watch_time, creator_tier, recency, history_match):
score, metrics = enhanced_value_model(
likes, comments, shares, saves, watch_time, creator_tier, recency, history_match
)
# Create metrics visualization
fig = go.Figure()
metric_names = list(metrics.keys())[:5] # First 5 are probabilities
metric_values = [metrics[name] for name in metric_names]
fig.add_trace(go.Bar(
x=metric_names,
y=metric_values,
marker_color=['#FF6B6B', '#4ECDC4', '#45B7D1', '#96CEB4', '#FECA57']
))
fig.update_layout(
title="Signal Probabilities",
yaxis_title="Probability",
height=400
)
text = f"""
### π’ Enhanced Value Model Output
**Final Ranking Score: {score}**
#### π Signal Probabilities:
- π Like: **{metrics['P(Like)']}**
- π¬ Comment: **{metrics['P(Comment)']}**
- π Share: **{metrics['P(Share)']}**
- π Save: **{metrics['P(Save)']}**
- π Watch: **{metrics['P(Watch)']}**
#### π― Modifiers:
- Creator Influence: **{metrics['Creator Multiplier']}**
- Content Freshness: **{metrics['Recency Factor']}**
- User History Match: **{metrics['History Match']}**
"""
return text, fig
def tab_enhanced_cold_start(interactions, topics, demographics, region, exploration):
user_vec = enhanced_cold_start(interactions, topics, demographics, region, exploration)
# Calculate distances to different content types
distances = {}
for topic, embedding in POST_EMBEDDINGS.items():
distance = np.linalg.norm(user_vec - embedding)
distances[topic] = distance
# Sort by proximity
sorted_topics = sorted(distances.items(), key=lambda x: x[1])
explanation = f"""
### π§ Enhanced Cold Start Analysis
**User Profile:**
- Interactions: **{interactions}**
- Demographics: **{demographics}** in **{region}**
- Exploration Factor: **{exploration}**
- Preferred Topics: **{', '.join(topics) if topics else 'None selected'}**
#### π― Content Affinity (Closest β Farthest):
"""
for i, (topic, dist) in enumerate(sorted_topics[:5]):
explanation += f"{i+1}. **{topic}** (distance: {dist:.3f})\n"
return explanation, user_vec
def tab_feed_ranking(user_vec):
if user_vec is None:
return "Please generate a user profile first in the Cold Start tab.", None
ranked_content = rank_content_feed(user_vec)
# Create feed visualization
df_feed = pd.DataFrame([
{
"Position": i+1,
"Topic": item["topic"],
"Creator": item["creator_tier"],
"Relevance": round(item["relevance"], 3),
"Value Score": round(item["value_score"], 3),
"Likes": round(item["likes"], 1),
"Comments": round(item["comments"], 1),
"Shares": round(item["shares"], 1)
}
for i, item in enumerate(ranked_content)
])
# Create ranking visualization
fig = px.scatter(
df_feed,
x="Relevance",
y="Value Score",
size="Likes",
color="Topic",
hover_data=["Creator", "Comments", "Shares"],
title="Content Ranking: Relevance vs Value Score"
)
return df_feed, fig
def tab_advanced_analytics(user_vec):
if user_vec is None:
return None, None, "Generate user profile first"
# Topic affinity radar chart
topic_scores = []
for topic, embedding in POST_EMBEDDINGS.items():
similarity = np.dot(user_vec, embedding) / (
np.linalg.norm(user_vec) * np.linalg.norm(embedding)
)
topic_scores.append(similarity)
fig_radar = go.Figure()
fig_radar.add_trace(go.Scatterpolar(
r=topic_scores,
theta=TOPICS,
fill='toself',
name='User Affinity'
))
fig_radar.update_layout(
polar=dict(
radialaxis=dict(
visible=True,
range=[-1, 1]
)),
showlegend=True,
title="User Topic Affinity Profile"
)
# Embedding visualization (PCA to 2D)
all_embeddings = list(POST_EMBEDDINGS.values()) + [user_vec]
all_labels = TOPICS + ["User"]
# Simple 2D projection (first two dimensions)
x_coords = [emb[0] for emb in all_embeddings]
y_coords = [emb[1] for emb in all_embeddings]
fig_embed = go.Figure()
# Plot topics
for i, (x, y, label) in enumerate(zip(x_coords[:-1], y_coords[:-1], all_labels[:-1])):
fig_embed.add_trace(go.Scatter(
x=[x], y=[y],
mode='markers+text',
text=[label],
textposition="top center",
marker=dict(size=10),
name=label
))
# Plot user
fig_embed.add_trace(go.Scatter(
x=[x_coords[-1]], y=[y_coords[-1]],
mode='markers+text',
text=["You"],
textposition="top center",
marker=dict(size=15, color='red'),
name="User"
))
fig_embed.update_layout(
title="2D Embedding Space Projection",
xaxis_title="Dimension 1",
yaxis_title="Dimension 2"
)
# Statistics
stats = f"""
### π Advanced Analytics
**User Vector Statistics:**
- Vector Magnitude: {np.linalg.norm(user_vec):.3f}
- Dominant Dimensions: {np.argmax(np.abs(user_vec))}, {np.argsort(np.abs(user_vec))[-2]}
- Diversity Score: {np.std(topic_scores):.3f}
- Global Alignment: {np.dot(user_vec, GLOBAL_EMBEDDING) / (np.linalg.norm(user_vec) * np.linalg.norm(GLOBAL_EMBEDDING)):.3f}
"""
return fig_radar, fig_embed, stats
# ---------------------------
# NAVIGATION FUNCTIONS
# ---------------------------
TAB_LABELS = {
0: "Next β‘ Cold Start",
1: "Next β‘ Feed Ranking",
2: "Next β‘ Analytics",
3: "π Complete!"
}
MAX_TAB = 3
def go_next(current_tab):
new_tab = min(current_tab + 1, MAX_TAB)
return new_tab
def go_prev(current_tab):
new_tab = max(current_tab - 1, 0)
return new_tab
def update_next_label(current_tab):
return TAB_LABELS.get(current_tab, "Next β‘")
def update_prev_visibility(current_tab):
return gr.update(visible=current_tab > 0)
# ---------------------------
# ENHANCED GRADIO UI
# ---------------------------
with gr.Blocks(theme=gr.themes.Soft()) as demo:
gr.Markdown("# πΈ Advanced Instagram Recommendation Algorithm Simulator")
gr.Markdown("### Explore the complex mechanics behind social media content ranking")
# Navigation state
current_tab = gr.State(0)
# Navigation buttons (moved outside tabs)
# Main tabs container
tabs = gr.Tabs(selected=0)
with tabs:
# ---------------- TAB A: Enhanced Value Model ----------------
with gr.Tab("π’ Value Model", id=0):
gr.Markdown("### Multi-Signal Content Scoring System")
with gr.Row():
with gr.Column():
likes = gr.Slider(0, 20, 5, label="π Likes Signal")
comments = gr.Slider(0, 10, 2, label="π¬ Comments Signal")
shares = gr.Slider(0, 5, 1, label="π Shares Signal")
saves = gr.Slider(0, 3, 0.5, label="π Saves Signal")
with gr.Column():
watch_time = gr.Slider(0, 15, 5, label="π Watch Time (seconds)")
creator_tier = gr.Dropdown(list(CREATOR_TIERS.keys()), value="Mid", label="π€ Creator Tier")
recency = gr.Slider(0.1, 48, 2, label="β° Hours Since Posted")
history_match = gr.Slider(0, 1, 0.5, label="π― User History Match")
value_output = gr.Markdown()
value_chart = gr.Plot()
gr.Button("π Calculate Ranking Score", variant="primary").click(
tab_enhanced_value_model,
inputs=[likes, comments, shares, saves, watch_time, creator_tier, recency, history_match],
outputs=[value_output, value_chart]
)
# ---------------- TAB B: Enhanced Cold Start ----------------
with gr.Tab("π§ Cold Start & Personalization", id=1):
gr.Markdown("### From Generic β Personalized Content")
with gr.Row():
with gr.Column():
interactions = gr.Slider(0, 100, 0, label="π± Total User Interactions")
topics = gr.CheckboxGroup(TOPICS, label="β€οΈ Preferred Topics")
with gr.Column():
demographics = gr.Dropdown(DEMOGRAPHICS, value="Gen Z", label="π₯ Demographics")
region = gr.Dropdown(REGIONS, value="North America", label="π Region")
exploration = gr.Slider(0, 0.5, 0.2, label="π² Exploration Factor")
cold_start_output = gr.Markdown()
user_vec_state = gr.State()
gr.Button("π Generate User Profile", variant="primary").click(
tab_enhanced_cold_start,
inputs=[interactions, topics, demographics, region, exploration],
outputs=[cold_start_output, user_vec_state]
)
# ---------------- TAB C: Feed Ranking ----------------
with gr.Tab("π± Feed Ranking Simulation", id=2):
gr.Markdown("### See Your Personalized Feed in Action")
feed_table = gr.Dataframe()
feed_chart = gr.Plot()
gr.Button("π Generate My Feed", variant="primary").click(
tab_feed_ranking,
inputs=[user_vec_state],
outputs=[feed_table, feed_chart]
)
# ---------------- TAB D: Advanced Analytics ----------------
with gr.Tab("π Advanced Analytics", id=3):
gr.Markdown("### Deep Dive into Algorithm Mechanics")
analytics_stats = gr.Markdown()
with gr.Row():
radar_chart = gr.Plot()
embedding_chart = gr.Plot()
gr.Button("π¬ Analyze User Profile", variant="primary").click(
tab_advanced_analytics,
inputs=[user_vec_state],
outputs=[radar_chart, embedding_chart, analytics_stats]
)
# ---------------------------
# TAB NAVIGATION LOGIC (FIXED)
# ---------------------------
with gr.Row():
nav_prev = gr.Button("β¬
Back", size="sm", visible=False)
nav_next = gr.Button("Next β‘ Cold Start", size="sm")
# Next button handler
nav_next.click(
fn=go_next,
inputs=current_tab,
outputs=current_tab
).then(
fn=lambda tab: gr.update(selected=tab),
inputs=current_tab,
outputs=tabs
).then(
fn=update_next_label,
inputs=current_tab,
outputs=nav_next
).then(
fn=update_prev_visibility,
inputs=current_tab,
outputs=nav_prev
)
# Previous button handler
nav_prev.click(
fn=go_prev,
inputs=current_tab,
outputs=current_tab
).then(
fn=lambda tab: gr.update(selected=tab),
inputs=current_tab,
outputs=tabs
).then(
fn=update_next_label,
inputs=current_tab,
outputs=nav_next
).then(
fn=update_prev_visibility,
inputs=current_tab,
outputs=nav_prev
)
# ---------------- Information Panel ----------------
with gr.Accordion("π Algorithm Insights", open=False):
gr.Markdown("""
### How This Simulation Works:
1. **Value Model**: Converts user engagement signals into probability scores using sigmoid functions
2. **Cold Start**: Blends global trends with personal preferences based on interaction history
3. **Embeddings**: Represents users and content in high-dimensional vector space
4. **Ranking**: Combines relevance scores with diversity penalties for balanced feeds
5. **Personalization**: Gradually shifts from trending to personalized content
### Key Concepts:
- **Attention Mechanism**: Weighted content selection based on user interests
- **Temporal Decay**: Newer content gets priority boost
- **Diversity Penalty**: Prevents echo chambers by promoting content variety
- **Demographic Targeting**: Adjusts recommendations based on user demographics
- **Exploration vs Exploitation**: Balance between showing familiar and new content
""")
with gr.Accordion("βοΈ Technical Implementation", open=False):
gr.Markdown("""
### Advanced Features:
- **8-Dimensional Embedding Space** for richer content representation
- **Multi-Signal Value Model** with 5 engagement types
- **Demographic & Regional Biases** in recommendation
- **Dynamic Exploration Factor** for content discovery
- **Attention-Based Ranking** with diversity constraints
- **Temporal Content Decay** for freshness prioritization
- **Creator Tier Influence** on engagement predictions
""")
if __name__ == "__main__":
demo.launch(debug=True) |