{ "cells": [ { "cell_type": "markdown", "id": "a669562d", "metadata": {}, "source": [ "# PageRank Approach for Finding Users with Good Taste\n", "\n", "## Plan:\n", "\n", "### 1. Build Directed Graph\n", "- Nodes: Users and Clips\n", "- Edges: User → Clip (if user likes clip, weight by log of play_count)\n", "- Edges: Clip → User (distribute clip's quality to users who liked it)\n", "\n", "### 2. Define \"Good Clips\" (Bot-Resistant)\n", "- **High unique user engagement**: likes/unique_users ratio\n", "- **High average play count**: mean(play_count) per user\n", "- **Low flag rate**: flagged_count/total_interactions\n", "- **Engagement diversity**: liked by users with diverse listening patterns\n", "- **NOT just high like count** (this attracts bots)\n", "\n", "### 3. Initialize Scores\n", "- Clips: Score based on unique user engagement metrics above\n", "- Users: \n", " - Users who like rarely and wegith their vote more.\n", " - for example, if they listened to 100 clips and only liked 4, each vote should get a weight of 25.\n", " - if they listened to 100 clips and liked 50, each vote should get a weight of 2 ish.\n", "\n", "### 4. Bot Detection & Filtering\n", "- Remove users who like >80% of clips they interact with\n", "- Remove users with avg play_count < 1 but many likes\n", "- Remove users who liked >100 clips but play_count ≈ 0\n", "- Weight edges by play_count to reduce bot influence\n", "\n", "### 5. Run PageRank\n", "- Users gain score from liking genuinely engaging clips\n", "- Clips gain score from being liked by high-score (non-bot) users\n", "- Iterate until convergence" ] }, { "cell_type": "code", "execution_count": 1, "id": "6c0e3a2d", "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "import numpy as np\n", "import networkx as nx\n", "from typing import Dict, List, Set, Tuple\n", "import matplotlib.pyplot as plt\n", "import seaborn as sns\n", "from collections import defaultdict\n", "import warnings\n", "\n", "warnings.filterwarnings(\"ignore\")\n", "\n", "# Load the data\n", "reactions_on_others = pd.read_pickle(\n", " \"/home/tony/Data/Preference/user_reaction_interaction_20250820.pkl\"\n", ")" ] }, { "cell_type": "code", "execution_count": 2, "id": "3eaa6658", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Data shape: (3329371, 10)\n", "\n", "Columns: ['clip_id', 'user_id', 'updated_at', 'play_count', 'skip_count', 'flagged', 'flagged_reason', 'reaction_type', 'is_pro_user', 'clip_creator_id']\n", "\n", "Reaction types: reaction_type\n", "L 1308021\n", "D 2520\n", "Name: count, dtype: int64\n", "\n", "Interactions by type:\n", " - With reaction (L or D): 1310541\n", " - No reaction (just listen): 2018830\n", " - Total interactions: 3329371\n", "\n", "Data types:\n", "user_id type: int32\n", "clip_id type: object\n" ] } ], "source": [ "# Check data structure\n", "print(\"Data shape:\", reactions_on_others.shape)\n", "print(\"\\nColumns:\", reactions_on_others.columns.tolist())\n", "print(\"\\nReaction types:\", reactions_on_others[\"reaction_type\"].value_counts())\n", "\n", "# KEY INSIGHT: Many interactions have no reaction (just listening)\n", "print(\"\\nInteractions by type:\")\n", "print(\n", " f\" - With reaction (L or D): {reactions_on_others['reaction_type'].notna().sum()}\"\n", ")\n", "print(\n", " f\" - No reaction (just listen): {reactions_on_others['reaction_type'].isna().sum()}\"\n", ")\n", "print(f\" - Total interactions: {len(reactions_on_others)}\")\n", "\n", "# Check data types for IDs\n", "print(\"\\nData types:\")\n", "print(f\"user_id type: {reactions_on_others['user_id'].dtype}\")\n", "print(f\"clip_id type: {reactions_on_others['clip_id'].dtype}\")" ] }, { "cell_type": "code", "execution_count": 3, "id": "0eaaf288", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Analyzing user behavior patterns...\n", "\n", "User behavior statistics (all 153972 users):\n", " - Mean like rate: 5.6%\n", " - Median like rate: 0.0%\n", " - Users with like_rate > 0.5: 5259\n", " - Users with like_rate > 0.7: 3250\n", " - Users with like_rate > 0.9: 1865\n", "\n", "Interaction distribution:\n", " - Users with 1-10 interactions: 118474\n", " - Users with 11-50 interactions: 27183\n", " - Users with 51+ interactions: 8315\n" ] } ], "source": [ "# Analyze user behavior patterns before bot detection\n", "print(\"Analyzing user behavior patterns...\")\n", "user_behavior = reactions_on_others.groupby(\"user_id\").agg(\n", " {\n", " \"clip_id\": [\"count\", \"nunique\"],\n", " \"reaction_type\": [\n", " lambda x: (x == \"L\").sum(),\n", " lambda x: x.isna().sum(),\n", " ],\n", " \"play_count\": \"mean\",\n", " }\n", ")\n", "\n", "user_behavior.columns = [\n", " \"total_interactions\",\n", " \"unique_clips\",\n", " \"likes\",\n", " \"neutral_listens\",\n", " \"avg_play_count\",\n", "]\n", "user_behavior[\"like_rate\"] = (\n", " user_behavior[\"likes\"] / user_behavior[\"total_interactions\"]\n", ")\n", "\n", "print(f\"\\nUser behavior statistics (all {len(user_behavior)} users):\")\n", "print(f\" - Mean like rate: {user_behavior['like_rate'].mean():.1%}\")\n", "print(f\" - Median like rate: {user_behavior['like_rate'].median():.1%}\")\n", "print(\n", " f\" - Users with like_rate > 0.5: {len(user_behavior[user_behavior['like_rate'] > 0.5])}\"\n", ")\n", "print(\n", " f\" - Users with like_rate > 0.7: {len(user_behavior[user_behavior['like_rate'] > 0.7])}\"\n", ")\n", "print(\n", " f\" - Users with like_rate > 0.9: {len(user_behavior[user_behavior['like_rate'] > 0.9])}\"\n", ")\n", "\n", "# Show distribution of interactions\n", "print(f\"\\nInteraction distribution:\")\n", "print(\n", " f\" - Users with 1-10 interactions: {len(user_behavior[user_behavior['total_interactions'] <= 10])}\"\n", ")\n", "print(\n", " f\" - Users with 11-50 interactions: {len(user_behavior[(user_behavior['total_interactions'] > 10) & (user_behavior['total_interactions'] <= 50)])}\"\n", ")\n", "print(\n", " f\" - Users with 51+ interactions: {len(user_behavior[user_behavior['total_interactions'] > 50])}\"\n", ")" ] }, { "cell_type": "code", "execution_count": 4, "id": "b4d3050a", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Interaction Breakdown:\n", " - Neutral listens: 2,018,830 (60.6%)\n", " - Likes: 1,308,021 (39.3%)\n", " - Dislikes: 2,520 (0.1%)\n", "\n", "This means 60.6% of all interactions were being ignored in the original analysis!\n" ] } ], "source": [ "# Analyze the significance of neutral listens\n", "total_interactions = len(reactions_on_others)\n", "neutral_listens = reactions_on_others[\"reaction_type\"].isna().sum()\n", "likes = (reactions_on_others[\"reaction_type\"] == \"L\").sum()\n", "dislikes = (reactions_on_others[\"reaction_type\"] == \"D\").sum()\n", "\n", "print(\"Interaction Breakdown:\")\n", "print(\n", " f\" - Neutral listens: {neutral_listens:,} ({neutral_listens/total_interactions:.1%})\"\n", ")\n", "print(f\" - Likes: {likes:,} ({likes/total_interactions:.1%})\")\n", "print(f\" - Dislikes: {dislikes:,} ({dislikes/total_interactions:.1%})\")\n", "print(\n", " f\"\\nThis means {neutral_listens/total_interactions:.1%} of all interactions were being ignored in the original analysis!\"\n", ")" ] }, { "cell_type": "code", "execution_count": 5, "id": "b46bffaf", "metadata": {}, "outputs": [], "source": [ "# # Debug: Check if specific high like-rate users were caught as bots\n", "# suspicious_user_ids = [75567253, 100442050, 65805460]\n", "# print(\"\\nDEBUG: Checking if suspicious users were detected as bots:\")\n", "# for uid in suspicious_user_ids:\n", "# is_bot = uid in bot_users\n", "# print(f\" - User {uid}: {'BOT' if is_bot else 'NOT BOT'}\")\n", "\n", "# # Check their stats in the original data\n", "# print(\"\\nTheir stats in original data:\")\n", "# user_stats_debug = reactions_on_others.groupby(\"user_id\").agg(\n", "# {\n", "# \"reaction_type\": [\n", "# \"count\",\n", "# lambda x: (x == \"L\").sum(),\n", "# ]\n", "# }\n", "# )\n", "# user_stats_debug.columns = [\"total_interactions\", \"likes\"]\n", "# user_stats_debug[\"like_rate\"] = (\n", "# user_stats_debug[\"likes\"] / user_stats_debug[\"total_interactions\"]\n", "# )\n", "\n", "# for uid in suspicious_user_ids:\n", "# if uid in user_stats_debug.index:\n", "# stats = user_stats_debug.loc[uid]\n", "# print(\n", "# f\" - User {uid}: {stats['total_interactions']} interactions, {stats['likes']} likes, {stats['like_rate']:.1%} like rate\"\n", "# )\n", "\n", "# # Check if they still exist in filtered data\n", "# print(\"\\nChecking if they exist in filtered data:\")\n", "# for uid in suspicious_user_ids:\n", "# count_in_filtered = len(df_filtered[df_filtered[\"user_id\"] == uid])\n", "# print(f\" - User {uid}: {count_in_filtered} interactions in filtered data\")\n", "\n", "# # Show bot detection thresholds\n", "# print(\"\\n\\nCurrent bot detection thresholds:\")\n", "# print(\" - Extreme likers: like_rate > 0.9 AND interactions > 5\")\n", "# print(\" - High like rate: like_rate > 0.6 AND interactions > 20\")\n", "# print(\" - Always react: reaction_rate > 0.8 AND interactions > 30\")\n", "# print(\" - Like everything: neutral_listens < likes * 0.5 AND likes > 30\")" ] }, { "cell_type": "code", "execution_count": 6, "id": "f5883f8e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Detected 4277 bot users:\n", " - High like rate (>80%): 4277\n", " - Many likes but no play: 0\n", " - High interact, low play: 0\n", " - No neutral listening: 436\n", "\n", "Bot vs Normal user statistics:\n", " - Bot like rate: 84.6% vs Normal: 3.3%\n", " - Bot avg play per like: 2.10 vs Normal: 68771168043.96\n", " - Bot neutral listens: 48 vs Normal: 12\n", "\n", "Filtered data shape: (2118572, 10) (removed 1210799 bot interactions)\n" ] } ], "source": [ "# Step 1: Bot Detection and Filtering\n", "def detect_bots(df: pd.DataFrame) -> Set[int]:\n", " \"\"\"Detect bot users based on suspicious behavior patterns.\"\"\"\n", " bots = set()\n", "\n", " # Calculate user metrics - ALL rows are interactions\n", " user_stats = df.groupby(\"user_id\").agg(\n", " {\n", " \"clip_id\": [\"count\", \"nunique\"], # total interactions and unique clips\n", " \"reaction_type\": [\n", " lambda x: (x == \"L\").sum(), # likes\n", " lambda x: (x == \"D\").sum(), # dislikes\n", " lambda x: x.isna().sum(), # listens without reaction\n", " ],\n", " \"play_count\": [\"mean\", \"sum\"],\n", " }\n", " )\n", "\n", " user_stats.columns = [\n", " \"total_interactions\",\n", " \"unique_clips\",\n", " \"likes\",\n", " \"dislikes\",\n", " \"neutral_listens\",\n", " \"avg_play_count\",\n", " \"total_play_count\",\n", " ]\n", "\n", " # Calculate metrics\n", " user_stats[\"like_rate\"] = user_stats[\"likes\"] / (\n", " user_stats[\"total_interactions\"] + 1e-10\n", " )\n", " user_stats[\"avg_play_per_like\"] = user_stats[\"total_play_count\"] / (\n", " user_stats[\"likes\"] + 1e-10\n", " )\n", "\n", " # Bot criteria - more aggressive filtering\n", " # 1. High like rate (>60% is abnormal given mean is 5.6%)\n", " high_like_rate = user_stats[\n", " (\n", " user_stats[\"like_rate\"] > 0.6\n", " ) # Lower threshold - normal users rarely like >60%\n", " ].index\n", " bots.update(high_like_rate)\n", "\n", " # 2. Many likes but almost no play count\n", " many_likes_no_play = user_stats[\n", " (user_stats[\"likes\"] > 100) & (user_stats[\"avg_play_per_like\"] < 0.1)\n", " ].index\n", " bots.update(many_likes_no_play)\n", "\n", " # 3. High interactions but low consumption\n", " high_interact_low_play = user_stats[\n", " (user_stats[\"total_interactions\"] > 100) & (user_stats[\"avg_play_count\"] < 1.0)\n", " ].index\n", " bots.update(high_interact_low_play)\n", "\n", " # 4. No neutral listening behavior (only likes, no exploration)\n", " no_neutral_listening = user_stats[\n", " (user_stats[\"neutral_listens\"] < user_stats[\"likes\"] * 0.2)\n", " & (user_stats[\"likes\"] > 50)\n", " ].index\n", " bots.update(no_neutral_listening)\n", "\n", " print(f\"Detected {len(bots)} bot users:\")\n", " print(f\" - High like rate (>80%): {len(high_like_rate)}\")\n", " print(f\" - Many likes but no play: {len(many_likes_no_play)}\")\n", " print(f\" - High interact, low play: {len(high_interact_low_play)}\")\n", " print(f\" - No neutral listening: {len(no_neutral_listening)}\")\n", "\n", " # Show some stats about detected bots vs normal users\n", " if len(bots) > 0:\n", " bot_stats = user_stats.loc[list(bots)]\n", " normal_stats = user_stats[~user_stats.index.isin(bots)]\n", " print(f\"\\nBot vs Normal user statistics:\")\n", " print(\n", " f\" - Bot like rate: {bot_stats['like_rate'].mean():.1%} vs Normal: {normal_stats['like_rate'].mean():.1%}\"\n", " )\n", " print(\n", " f\" - Bot avg play per like: {bot_stats['avg_play_per_like'].mean():.2f} vs Normal: {normal_stats['avg_play_per_like'].mean():.2f}\"\n", " )\n", " print(\n", " f\" - Bot neutral listens: {bot_stats['neutral_listens'].mean():.0f} vs Normal: {normal_stats['neutral_listens'].mean():.0f}\"\n", " )\n", "\n", " return bots\n", "\n", "\n", "# Detect and filter bots\n", "bot_users = detect_bots(reactions_on_others)\n", "df_filtered = reactions_on_others[\n", " ~reactions_on_others[\"user_id\"].isin(bot_users)\n", "].copy()\n", "print(\n", " f\"\\nFiltered data shape: {df_filtered.shape} (removed {len(reactions_on_others) - len(df_filtered)} bot interactions)\"\n", ")" ] }, { "cell_type": "code", "execution_count": 7, "id": "7ce731fb", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Analyzing clip popularity after bot filtering...\n", "Distribution of unique likers per clip:\n", " - 1 unique liker(s): 103353 clips\n", " - 2 unique liker(s): 20463 clips\n", " - 3 unique liker(s): 8880 clips\n", " - 4 unique liker(s): 4802 clips\n", " - 5 unique liker(s): 2831 clips\n", " - 6 unique liker(s): 1887 clips\n", " - 7 unique liker(s): 1259 clips\n", " - 8 unique liker(s): 871 clips\n", " - 9 unique liker(s): 693 clips\n", " - 10 unique liker(s): 518 clips\n", " - 11+ unique likers: 2786 clips\n", " - Max unique likers: 304\n" ] } ], "source": [ "# Analyze clip popularity after bot filtering\n", "print(\"\\nAnalyzing clip popularity after bot filtering...\")\n", "likes_only = df_filtered[df_filtered[\"reaction_type\"] == \"L\"]\n", "clip_popularity = (\n", " likes_only.groupby(\"clip_id\")[\"user_id\"].nunique().value_counts().sort_index()\n", ")\n", "\n", "print(f\"Distribution of unique likers per clip:\")\n", "for n_likers in range(1, min(11, len(clip_popularity) + 1)):\n", " if n_likers in clip_popularity.index:\n", " print(f\" - {n_likers} unique liker(s): {clip_popularity[n_likers]} clips\")\n", "\n", "if clip_popularity.index.max() > 10:\n", " print(\n", " f\" - 11+ unique likers: {clip_popularity[clip_popularity.index > 10].sum()} clips\"\n", " )\n", " print(f\" - Max unique likers: {clip_popularity.index.max()}\")" ] }, { "cell_type": "code", "execution_count": 8, "id": "8b8d6f46", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "User Selectivity Weight Statistics:\n", " - Total users: 149695\n", " - Mean selectivity weight: 26.26\n", " - Median selectivity weight: 20.00\n", " - Max selectivity weight: 50.00\n", " - Users with weight > 10: 92416\n", " - Users with weight > 20: 67465\n", "\n", "Examples of user selectivity:\n", "Most selective users (high weight):\n", " total_interactions total_likes selectivity_weight like_rate\n", "user_id \n", "726 16 0 50.0 0.0\n", "1878 7 0 50.0 0.0\n", "2011 16 0 50.0 0.0\n", "2205 10 0 50.0 0.0\n", "2214 19 0 50.0 0.0\n", "\n", "Least selective users (low weight):\n", " total_interactions total_likes selectivity_weight like_rate\n", "user_id \n", "115307 5 3 1.612903 0.6\n", "211326 5 3 1.612903 0.6\n", "964009 5 3 1.612903 0.6\n", "1506566 5 3 1.612903 0.6\n", "4740857 5 3 1.612903 0.6\n", "\n", "Calculated selectivity weights for 149695 users\n" ] } ], "source": [ "# Calculate user selectivity weights for PageRank\n", "def calculate_user_weights(df: pd.DataFrame) -> pd.DataFrame:\n", " \"\"\"\n", " Calculate selectivity weights for each user based on their like behavior.\n", " Selective users (who like rarely) get higher weights.\n", "\n", " Weight = total_interactions / total_likes\n", " Example: 100 interactions, 4 likes → weight = 25\n", " Example: 100 interactions, 50 likes → weight = 2\n", " \"\"\"\n", " user_weights = df.groupby(\"user_id\").agg(\n", " {\n", " \"clip_id\": \"count\", # total interactions\n", " \"reaction_type\": lambda x: (x == \"L\").sum(), # total likes\n", " }\n", " )\n", "\n", " user_weights.columns = [\"total_interactions\", \"total_likes\"]\n", "\n", " # Calculate selectivity weight\n", " # Add small epsilon to avoid division by zero\n", " user_weights[\"selectivity_weight\"] = user_weights[\"total_interactions\"] / (\n", " user_weights[\"total_likes\"] + 0.1\n", " )\n", "\n", " # Cap weights to avoid extreme values\n", " # Max weight of 50 means even very selective users don't dominate completely\n", " user_weights[\"selectivity_weight\"] = user_weights[\"selectivity_weight\"].clip(\n", " upper=50\n", " )\n", "\n", " # Also calculate like rate for reference\n", " user_weights[\"like_rate\"] = (\n", " user_weights[\"total_likes\"] / user_weights[\"total_interactions\"]\n", " )\n", "\n", " print(\"User Selectivity Weight Statistics:\")\n", " print(f\" - Total users: {len(user_weights)}\")\n", " print(\n", " f\" - Mean selectivity weight: {user_weights['selectivity_weight'].mean():.2f}\"\n", " )\n", " print(\n", " f\" - Median selectivity weight: {user_weights['selectivity_weight'].median():.2f}\"\n", " )\n", " print(f\" - Max selectivity weight: {user_weights['selectivity_weight'].max():.2f}\")\n", " print(\n", " f\" - Users with weight > 10: {len(user_weights[user_weights['selectivity_weight'] > 10])}\"\n", " )\n", " print(\n", " f\" - Users with weight > 20: {len(user_weights[user_weights['selectivity_weight'] > 20])}\"\n", " )\n", "\n", " # Show examples of selective vs non-selective users\n", " print(\"\\nExamples of user selectivity:\")\n", " print(\"Most selective users (high weight):\")\n", " top_selective = user_weights.nlargest(5, \"selectivity_weight\")\n", " print(\n", " top_selective[\n", " [\"total_interactions\", \"total_likes\", \"selectivity_weight\", \"like_rate\"]\n", " ]\n", " )\n", "\n", " print(\"\\nLeast selective users (low weight):\")\n", " bottom_selective = user_weights.nsmallest(5, \"selectivity_weight\")\n", " print(\n", " bottom_selective[\n", " [\"total_interactions\", \"total_likes\", \"selectivity_weight\", \"like_rate\"]\n", " ]\n", " )\n", "\n", " return user_weights\n", "\n", "\n", "# Calculate user weights after bot filtering\n", "user_weights = calculate_user_weights(df_filtered)\n", "print(f\"\\nCalculated selectivity weights for {len(user_weights)} users\")" ] }, { "cell_type": "code", "execution_count": 9, "id": "d3d66032", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Clip quality metrics calculated with weighted likes.\n", "\n", "Clip statistics:\n", " - Total clips: 525265\n", " - Clips with 0 likes: 376922\n", " - Clips with 1-2 likes: 123816\n", " - Clips with 3+ likes: 24527\n", " - Clips with enough unique likers (>=3): 24527\n", "\n", "Impact of selectivity weighting:\n", " - Mean weight per like: 3.93\n", " - Median weight per like: 2.61\n", " - Max weight amplification: 50.00x\n", "\n", "Top 10 highest quality clips (with 3+ unique likers):\n", " raw_likes weighted_likes \\\n", "clip_id \n", "e33e8ae1-cd85-4ee4-9618-f2f7f9275744 304 1907.391646 \n", "249ac1f8-991b-47f1-98d8-e916c0efd84b 242 1360.541376 \n", "0c8fc5be-8895-4ebe-a3f4-5d0f2d2adf66 175 1360.586210 \n", "20de0a6d-b89e-45a1-8c1d-7030c602fdd2 144 1218.807056 \n", "b8ce68e5-e36b-48ab-bd73-364eb878fc61 112 1184.624732 \n", "d03ed0c4-29ea-4a4f-8923-5f8aea1b926b 137 1144.951783 \n", "be52d814-b25c-4759-92f8-1737599355c3 149 1121.178632 \n", "671b0482-19e5-4d99-b962-2482d021bc0f 155 1069.817867 \n", "57654b51-0f61-4d00-8ed0-e8cfabc2f621 174 1037.222012 \n", "444b6eec-d131-4b6b-aeae-952be273eb85 131 1036.530994 \n", "\n", " unique_likers unique_users \\\n", "clip_id \n", "e33e8ae1-cd85-4ee4-9618-f2f7f9275744 304.0 2927 \n", "249ac1f8-991b-47f1-98d8-e916c0efd84b 242.0 2867 \n", "0c8fc5be-8895-4ebe-a3f4-5d0f2d2adf66 175.0 4424 \n", "20de0a6d-b89e-45a1-8c1d-7030c602fdd2 144.0 1920 \n", "b8ce68e5-e36b-48ab-bd73-364eb878fc61 112.0 2565 \n", "d03ed0c4-29ea-4a4f-8923-5f8aea1b926b 137.0 4326 \n", "be52d814-b25c-4759-92f8-1737599355c3 149.0 3698 \n", "671b0482-19e5-4d99-b962-2482d021bc0f 155.0 2080 \n", "57654b51-0f61-4d00-8ed0-e8cfabc2f621 174.0 2037 \n", "444b6eec-d131-4b6b-aeae-952be273eb85 131.0 4134 \n", "\n", " like_rate_per_user avg_play_count \\\n", "clip_id \n", "e33e8ae1-cd85-4ee4-9618-f2f7f9275744 0.103825 1.376495 \n", "249ac1f8-991b-47f1-98d8-e916c0efd84b 0.084379 1.352633 \n", "0c8fc5be-8895-4ebe-a3f4-5d0f2d2adf66 0.039548 1.286618 \n", "20de0a6d-b89e-45a1-8c1d-7030c602fdd2 0.074961 1.313021 \n", "b8ce68e5-e36b-48ab-bd73-364eb878fc61 0.043648 1.245614 \n", "d03ed0c4-29ea-4a4f-8923-5f8aea1b926b 0.031662 1.254739 \n", "be52d814-b25c-4759-92f8-1737599355c3 0.040281 1.339913 \n", "671b0482-19e5-4d99-b962-2482d021bc0f 0.074483 1.258173 \n", "57654b51-0f61-4d00-8ed0-e8cfabc2f621 0.085378 1.275896 \n", "444b6eec-d131-4b6b-aeae-952be273eb85 0.031681 1.244557 \n", "\n", " quality_score_norm \n", "clip_id \n", "e33e8ae1-cd85-4ee4-9618-f2f7f9275744 1.000000 \n", "249ac1f8-991b-47f1-98d8-e916c0efd84b 0.713488 \n", "0c8fc5be-8895-4ebe-a3f4-5d0f2d2adf66 0.713400 \n", "20de0a6d-b89e-45a1-8c1d-7030c602fdd2 0.639166 \n", "b8ce68e5-e36b-48ab-bd73-364eb878fc61 0.621214 \n", "d03ed0c4-29ea-4a4f-8923-5f8aea1b926b 0.600427 \n", "be52d814-b25c-4759-92f8-1737599355c3 0.588126 \n", "671b0482-19e5-4d99-b962-2482d021bc0f 0.561158 \n", "57654b51-0f61-4d00-8ed0-e8cfabc2f621 0.544008 \n", "444b6eec-d131-4b6b-aeae-952be273eb85 0.543615 \n" ] } ], "source": [ "# Step 2: Calculate Bot-Resistant Clip Quality Metrics with Weighted Likes\n", "def calculate_clip_quality(\n", " df: pd.DataFrame, user_weights: pd.DataFrame\n", ") -> pd.DataFrame:\n", " \"\"\"\n", " Calculate quality metrics for clips using weighted likes from selective users.\n", "\n", " Key insight: A like from a selective user (high weight) is worth more than\n", " a like from someone who likes everything (low weight).\n", " \"\"\"\n", "\n", " # Get unique users per clip\n", " unique_users_per_clip = df.groupby(\"clip_id\")[\"user_id\"].nunique()\n", " likes_df = df[df[\"reaction_type\"] == \"L\"]\n", " unique_likers = likes_df.groupby(\"clip_id\")[\"user_id\"].nunique()\n", "\n", " # Calculate WEIGHTED likes - this is the key improvement\n", " weighted_likes_by_clip = {}\n", " for clip_id, clip_likes in likes_df.groupby(\"clip_id\"):\n", " weighted_sum = 0\n", " for _, like in clip_likes.iterrows():\n", " user_id = like[\"user_id\"]\n", " if user_id in user_weights.index:\n", " # Use selectivity weight for this user\n", " weight = user_weights.loc[user_id, \"selectivity_weight\"]\n", " else:\n", " weight = 1.0 # Default weight\n", " weighted_sum += weight\n", " weighted_likes_by_clip[clip_id] = weighted_sum\n", "\n", " # Convert to Series\n", " weighted_likes = pd.Series(weighted_likes_by_clip)\n", "\n", " # Aggregate metrics\n", " clip_metrics = df.groupby(\"clip_id\").agg(\n", " {\n", " \"user_id\": \"count\", # total interactions\n", " \"reaction_type\": [\n", " lambda x: (x == \"L\").sum(), # total likes (raw count)\n", " lambda x: (x == \"D\").sum(), # total dislikes\n", " ],\n", " \"play_count\": [\"mean\", \"sum\"],\n", " \"flagged\": \"sum\",\n", " }\n", " )\n", "\n", " clip_metrics.columns = [\n", " \"total_interactions\",\n", " \"raw_likes\", # Changed from 'likes' to 'raw_likes' for clarity\n", " \"dislikes\",\n", " \"avg_play_count\",\n", " \"total_play_count\",\n", " \"flagged_count\",\n", " ]\n", "\n", " # Add metrics\n", " clip_metrics[\"unique_likers\"] = unique_likers.fillna(0)\n", " clip_metrics[\"unique_users\"] = unique_users_per_clip\n", " clip_metrics[\"weighted_likes\"] = weighted_likes.fillna(0) # NEW: weighted likes\n", "\n", " # Calculate bot-resistant metrics\n", " clip_metrics[\"like_rate_per_user\"] = clip_metrics[\"raw_likes\"] / (\n", " clip_metrics[\"unique_users\"] + 1\n", " )\n", " clip_metrics[\"avg_play_per_liker\"] = clip_metrics[\"total_play_count\"] / (\n", " clip_metrics[\"raw_likes\"] + 1\n", " )\n", " clip_metrics[\"flag_rate\"] = clip_metrics[\"flagged_count\"] / (\n", " clip_metrics[\"total_interactions\"] + 1\n", " )\n", "\n", " # Calculate engagement diversity (std of play counts)\n", " engagement_diversity = df.groupby(\"clip_id\")[\"play_count\"].std()\n", " clip_metrics[\"engagement_diversity\"] = engagement_diversity.fillna(0)\n", "\n", " # Filter out clips with too few likes to be meaningful\n", " clip_metrics[\"has_enough_likes\"] = clip_metrics[\"unique_likers\"] >= 3\n", "\n", " # Composite quality score using WEIGHTED likes\n", " # Now selective users' likes contribute more to quality\n", " clip_metrics[\"quality_score\"] = (\n", " clip_metrics[\"weighted_likes\"]\n", " * 0.4 # Weighted likes from selective users (KEY CHANGE)\n", " + clip_metrics[\"like_rate_per_user\"] * 0.2 # % of listeners who liked it\n", " + clip_metrics[\"avg_play_count\"] * 0.2 # Real engagement\n", " + clip_metrics[\"engagement_diversity\"] * 0.1 # Diverse audience\n", " + (1 - clip_metrics[\"flag_rate\"]) * 0.1 # Not problematic\n", " )\n", "\n", " # Set quality to 0 for clips without enough likes\n", " clip_metrics.loc[~clip_metrics[\"has_enough_likes\"], \"quality_score\"] = 0\n", "\n", " # Normalize quality score\n", " min_score = clip_metrics[\"quality_score\"].min()\n", " max_score = clip_metrics[\"quality_score\"].max()\n", " clip_metrics[\"quality_score_norm\"] = (clip_metrics[\"quality_score\"] - min_score) / (\n", " max_score - min_score + 1e-10\n", " )\n", "\n", " return clip_metrics\n", "\n", "\n", "# Calculate clip quality WITH user weights\n", "clip_quality = calculate_clip_quality(df_filtered, user_weights)\n", "print(\"Clip quality metrics calculated with weighted likes.\")\n", "\n", "# Show distribution of likes\n", "print(f\"\\nClip statistics:\")\n", "print(f\" - Total clips: {len(clip_quality)}\")\n", "print(f\" - Clips with 0 likes: {len(clip_quality[clip_quality['raw_likes'] == 0])}\")\n", "print(\n", " f\" - Clips with 1-2 likes: {len(clip_quality[(clip_quality['raw_likes'] >= 1) & (clip_quality['raw_likes'] <= 2)])}\"\n", ")\n", "print(f\" - Clips with 3+ likes: {len(clip_quality[clip_quality['raw_likes'] >= 3])}\")\n", "print(\n", " f\" - Clips with enough unique likers (>=3): {len(clip_quality[clip_quality['has_enough_likes']])}\"\n", ")\n", "\n", "# Show impact of weighting\n", "print(f\"\\nImpact of selectivity weighting:\")\n", "liked_clips = clip_quality[clip_quality[\"raw_likes\"] > 0]\n", "if len(liked_clips) > 0:\n", " weight_ratio = liked_clips[\"weighted_likes\"] / liked_clips[\"raw_likes\"]\n", " print(f\" - Mean weight per like: {weight_ratio.mean():.2f}\")\n", " print(f\" - Median weight per like: {weight_ratio.median():.2f}\")\n", " print(f\" - Max weight amplification: {weight_ratio.max():.2f}x\")\n", "\n", "print(f\"\\nTop 10 highest quality clips (with 3+ unique likers):\")\n", "top_quality_clips = clip_quality[clip_quality[\"has_enough_likes\"]].nlargest(\n", " 10, \"quality_score_norm\"\n", ")\n", "if len(top_quality_clips) > 0:\n", " print(\n", " top_quality_clips[\n", " [\n", " \"raw_likes\",\n", " \"weighted_likes\",\n", " \"unique_likers\",\n", " \"unique_users\",\n", " \"like_rate_per_user\",\n", " \"avg_play_count\",\n", " \"quality_score_norm\",\n", " ]\n", " ]\n", " )\n", "else:\n", " print(\"No clips found with 3+ unique likers!\")" ] }, { "cell_type": "code", "execution_count": 10, "id": "f8a44973", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Building collaborative filtering graph...\n", "Building graph with 22959 users and 148343 clips\n", "Graph has 171302 nodes and 607896 edges\n", " - User nodes: 22959\n", " - Clip nodes: 148343\n", " - Avg clips liked per user: 13.2\n", " - Avg users per clip: 2.0\n" ] } ], "source": [ "# Step 3: Simplified Collaborative Filtering Graph\n", "def build_collaborative_graph(\n", " df: pd.DataFrame, user_weights: pd.DataFrame\n", ") -> nx.DiGraph:\n", " \"\"\"\n", " Build a simplified bipartite graph for collaborative filtering.\n", "\n", " Core idea: A user's taste score comes from liking clips that other selective users also like.\n", " - Selective users (high weight) strongly boost clips when they like them\n", " - Users who like highly-boosted clips get higher taste scores\n", " \"\"\"\n", " G = nx.DiGraph()\n", "\n", " # Only consider likes for collaborative filtering\n", " likes_df = df[df[\"reaction_type\"] == \"L\"]\n", "\n", " # Get unique users and clips that have likes\n", " users_with_likes = likes_df[\"user_id\"].unique()\n", " clips_with_likes = likes_df[\"clip_id\"].unique()\n", "\n", " print(\n", " f\"Building graph with {len(users_with_likes)} users and {len(clips_with_likes)} clips\"\n", " )\n", "\n", " # Add nodes\n", " for uid in users_with_likes:\n", " G.add_node(f\"u_{uid}\", bipartite=0, type=\"user\")\n", " for cid in clips_with_likes:\n", " G.add_node(f\"c_{cid}\", bipartite=1, type=\"clip\")\n", "\n", " # Build edges\n", " for _, like in likes_df.iterrows():\n", " user_id = like[\"user_id\"]\n", " clip_id = like[\"clip_id\"]\n", " user_node = f\"u_{user_id}\"\n", " clip_node = f\"c_{clip_id}\"\n", "\n", " # Get user's selectivity weight\n", " if user_id in user_weights.index:\n", " weight = user_weights.loc[user_id, \"selectivity_weight\"]\n", " else:\n", " weight = 1.0\n", "\n", " # User -> Clip: selective users boost clip importance\n", " G.add_edge(user_node, clip_node, weight=weight)\n", "\n", " # Clip -> User: clips pass score back, but INVERSELY weighted by selectivity\n", " # This ensures selective users get MORE benefit from good clips\n", " # Users who like everything get LESS benefit (diluted across too many clips)\n", " clip_to_user_weight = weight / 10.0 # Higher selectivity = higher return flow\n", " G.add_edge(clip_node, user_node, weight=clip_to_user_weight)\n", "\n", " return G\n", "\n", "\n", "# Build the collaborative filtering graph\n", "print(\"\\nBuilding collaborative filtering graph...\")\n", "cf_graph = build_collaborative_graph(df_filtered, user_weights)\n", "print(\n", " f\"Graph has {cf_graph.number_of_nodes()} nodes and {cf_graph.number_of_edges()} edges\"\n", ")\n", "\n", "# Show some statistics\n", "user_nodes = [n for n in cf_graph.nodes() if n.startswith(\"u_\")]\n", "clip_nodes = [n for n in cf_graph.nodes() if n.startswith(\"c_\")]\n", "print(f\" - User nodes: {len(user_nodes)}\")\n", "print(f\" - Clip nodes: {len(clip_nodes)}\")\n", "\n", "# Check connectivity\n", "avg_edges_per_user = sum(cf_graph.out_degree(u) for u in user_nodes) / len(user_nodes)\n", "avg_edges_per_clip = sum(cf_graph.out_degree(c) for c in clip_nodes) / len(clip_nodes)\n", "print(f\" - Avg clips liked per user: {avg_edges_per_user:.1f}\")\n", "print(f\" - Avg users per clip: {avg_edges_per_clip:.1f}\")" ] }, { "cell_type": "code", "execution_count": 11, "id": "dba36132", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Running PageRank for collaborative filtering...\n", "PageRank completed!\n", "\n", "Score distribution:\n", " - User scores: min=0.000001, max=0.003011\n", " - Clip scores: min=0.000001, max=0.000548\n" ] } ], "source": [ "# Step 4: Run PageRank for Collaborative Filtering\n", "print(\"\\nRunning PageRank for collaborative filtering...\")\n", "\n", "# Simple initialization - let the algorithm find taste through connections\n", "# PageRank will naturally flow from selective users (high weight edges) to clips they like,\n", "# and then to other users who like those same clips\n", "pagerank_scores = nx.pagerank(\n", " cf_graph,\n", " alpha=0.85, # damping factor\n", " weight=\"weight\", # use edge weights\n", " max_iter=100,\n", " tol=1e-06,\n", ")\n", "\n", "print(\"PageRank completed!\")\n", "\n", "# Analyze score distribution\n", "user_scores = {\n", " node: score for node, score in pagerank_scores.items() if node.startswith(\"u_\")\n", "}\n", "clip_scores = {\n", " node: score for node, score in pagerank_scores.items() if node.startswith(\"c_\")\n", "}\n", "\n", "print(f\"\\nScore distribution:\")\n", "print(\n", " f\" - User scores: min={min(user_scores.values()):.6f}, max={max(user_scores.values()):.6f}\"\n", ")\n", "print(\n", " f\" - Clip scores: min={min(clip_scores.values()):.6f}, max={max(clip_scores.values()):.6f}\"\n", ")" ] }, { "cell_type": "code", "execution_count": 12, "id": "2d005acf", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Top clips by PageRank score (these should be liked by selective users):\n", "\n", "Top 10 clips:\n", " clip_id pagerank_score num_likers \\\n", "42 e33e8ae1-cd85-4ee4-9618-f2f7f9275744 0.000548 304 \n", "551 249ac1f8-991b-47f1-98d8-e916c0efd84b 0.000408 242 \n", "163 0c8fc5be-8895-4ebe-a3f4-5d0f2d2adf66 0.000366 175 \n", "3721 6636284a-c63a-4994-9077-0cdb98d349e4 0.000323 200 \n", "1094 be52d814-b25c-4759-92f8-1737599355c3 0.000310 149 \n", "655 57654b51-0f61-4d00-8ed0-e8cfabc2f621 0.000305 174 \n", "6646 20de0a6d-b89e-45a1-8c1d-7030c602fdd2 0.000302 144 \n", "74 d03ed0c4-29ea-4a4f-8923-5f8aea1b926b 0.000298 137 \n", "2513 adb745e4-835e-44f0-8775-7c22ef0dac33 0.000297 186 \n", "3548 671b0482-19e5-4d99-b962-2482d021bc0f 0.000293 155 \n", "\n", " total_weight avg_weight \n", "42 1907.391646 6.274315 \n", "551 1360.541376 5.622072 \n", "163 1360.586210 7.774778 \n", "3721 977.700749 4.888504 \n", "1094 1121.178632 7.524689 \n", "655 1037.222012 5.961046 \n", "6646 1218.807056 8.463938 \n", "74 1144.951783 8.357312 \n", "2513 932.395550 5.012879 \n", "3548 1069.817867 6.902051 \n", "\n", "\n", "Checking if selective users' likes boost clips:\n", "\n", "Clips liked by highly selective users (avg weight > 10):\n", " clip_id pagerank_score num_likers \\\n", "170 b8ce68e5-e36b-48ab-bd73-364eb878fc61 0.000273 112 \n", "9598 814f6a8e-dc37-4b3b-92c3-cf6ced5d3707 0.000091 35 \n", "31974 f2748eba-f40f-4c34-9aec-991c88003c03 0.000058 24 \n", "25039 a1416c02-46e6-47ca-b073-0a0a57eb653c 0.000052 20 \n", "16158 71d045df-22f4-4801-8b93-7d9c77fd7247 0.000050 19 \n", "1678 90f6fca1-bfa2-42cb-aee3-8158e1674b11 0.000046 18 \n", "13616 b278585f-115c-4c3a-aff6-65d9cf67426a 0.000042 17 \n", "20572 b3d2d1d6-152e-41dc-bd5b-63d851a23fe8 0.000040 18 \n", "1110 62bf3ade-e015-4bb0-99e1-d74ea0859bb8 0.000040 15 \n", "30726 25add1d3-7d3c-48dc-8cfa-c5cdfefa637e 0.000039 13 \n", "\n", " avg_weight \n", "170 10.577007 \n", "9598 10.704945 \n", "31974 10.942087 \n", "25039 11.285176 \n", "16158 12.578871 \n", "1678 11.566355 \n", "13616 10.053759 \n", "20572 10.771132 \n", "1110 11.142277 \n", "30726 12.733145 \n" ] } ], "source": [ "# Analyze which clips got highest scores (should be those liked by selective users)\n", "print(\"\\nTop clips by PageRank score (these should be liked by selective users):\")\n", "\n", "# Get clip scores with metadata\n", "clip_analysis = []\n", "for clip_node, score in clip_scores.items():\n", " clip_id = clip_node[2:]\n", "\n", " # Get who liked this clip and their weights\n", " likers = []\n", " total_weight = 0\n", " for edge in cf_graph.in_edges(clip_node, data=True):\n", " user_node = edge[0]\n", " user_id = int(user_node[2:])\n", " weight = edge[2][\"weight\"]\n", " likers.append((user_id, weight))\n", " total_weight += weight\n", "\n", " clip_analysis.append(\n", " {\n", " \"clip_id\": clip_id,\n", " \"pagerank_score\": score,\n", " \"num_likers\": len(likers),\n", " \"total_weight\": total_weight,\n", " \"avg_weight\": total_weight / len(likers) if likers else 0,\n", " }\n", " )\n", "\n", "# Convert to DataFrame and sort\n", "clip_analysis_df = pd.DataFrame(clip_analysis)\n", "clip_analysis_df = clip_analysis_df.sort_values(\"pagerank_score\", ascending=False)\n", "\n", "print(\"\\nTop 10 clips:\")\n", "print(\n", " clip_analysis_df.head(10)[\n", " [\"clip_id\", \"pagerank_score\", \"num_likers\", \"total_weight\", \"avg_weight\"]\n", " ]\n", ")\n", "\n", "# Check correlation between selectivity weight and PageRank score\n", "print(\"\\n\\nChecking if selective users' likes boost clips:\")\n", "high_weight_clips = clip_analysis_df[clip_analysis_df[\"avg_weight\"] > 10].head(10)\n", "print(f\"\\nClips liked by highly selective users (avg weight > 10):\")\n", "print(high_weight_clips[[\"clip_id\", \"pagerank_score\", \"num_likers\", \"avg_weight\"]])" ] }, { "cell_type": "code", "execution_count": 13, "id": "dfba4ebc", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Total users with taste scores: 22959\n", "\n", "DEBUG: Investigating high like-rate users in results...\n", "Found 0 users with like_rate > 0.9 in the results\n", "\n", "Top 20 users with best taste:\n", " user_id taste_score total_interactions total_likes like_rate \\\n", "228 78364115 1.000000 6101 3653 0.598754 \n", "613 89271156 0.940431 6513 3380 0.518962 \n", "692 75567253 0.910769 7209 2128 0.295187 \n", "287 62795185 0.879819 5278 2818 0.533914 \n", "344 72663343 0.850619 4804 2428 0.505412 \n", "522 1173818 0.635328 4832 1451 0.300290 \n", "304 88712285 0.526536 3062 1343 0.438602 \n", "240 96883241 0.523078 4014 848 0.211261 \n", "668 5515095 0.491851 3300 1436 0.435152 \n", "93 28605241 0.441019 3308 1729 0.522672 \n", "1043 86854140 0.435734 3571 1109 0.310557 \n", "411 65956664 0.420605 2997 1532 0.511178 \n", "1648 6012821 0.419117 2956 1393 0.471245 \n", "2 100990846 0.416339 2449 1356 0.553695 \n", "201 95915360 0.415847 2877 1584 0.550574 \n", "250 86499376 0.387642 2278 1211 0.531607 \n", "703 85309214 0.375416 3069 861 0.280547 \n", "666 65961470 0.373401 2158 1179 0.546339 \n", "33 63392178 0.371392 3232 1452 0.449257 \n", "1013 76055629 0.369239 1847 1077 0.583108 \n", "\n", " avg_play_count unique_clips \n", "228 1.165711 6101 \n", "613 1.445263 6513 \n", "692 1.287696 7209 \n", "287 1.709170 5278 \n", "344 1.201499 4804 \n", "522 1.152318 4832 \n", "304 1.183867 3062 \n", "240 1.059791 4014 \n", "668 1.115152 3300 \n", "93 1.174123 3308 \n", "1043 1.385046 3571 \n", "411 1.275609 2997 \n", "1648 1.502368 2956 \n", "2 1.265006 2449 \n", "201 1.278415 2877 \n", "250 1.165496 2278 \n", "703 1.327794 3069 \n", "666 1.132530 2158 \n", "33 1.626238 3232 \n", "1013 1.078506 1847 \n" ] } ], "source": [ "# Step 5: Extract and Analyze User Taste Scores\n", "def extract_user_taste_scores(\n", " pagerank_scores: Dict, df: pd.DataFrame, min_interactions: int = 1\n", ") -> pd.DataFrame:\n", " \"\"\"Extract user taste scores from PageRank results.\"\"\"\n", "\n", " # Extract user scores\n", " user_scores = {}\n", " for node, score in pagerank_scores.items():\n", " if node.startswith(\"u_\"):\n", " user_id = int(node[2:])\n", " user_scores[user_id] = score\n", "\n", " # Convert to dataframe\n", " taste_df = pd.DataFrame(\n", " list(user_scores.items()), columns=[\"user_id\", \"pagerank_score\"]\n", " )\n", "\n", " # Add user statistics\n", " # FIX: Count ALL rows, not just non-null reaction_type\n", " user_stats = df.groupby(\"user_id\").agg(\n", " {\n", " \"clip_id\": \"count\", # total interactions (counts all rows)\n", " \"reaction_type\": lambda x: (x == \"L\").sum(), # likes\n", " \"play_count\": \"mean\",\n", " }\n", " )\n", "\n", " user_stats.columns = [\n", " \"total_interactions\",\n", " \"total_likes\",\n", " \"avg_play_count\",\n", " ]\n", "\n", " # Add unique clips count separately\n", " unique_clips = df.groupby(\"user_id\")[\"clip_id\"].nunique()\n", " user_stats[\"unique_clips\"] = unique_clips\n", "\n", " # Calculate like rate to identify potential bots that slipped through\n", " user_stats[\"like_rate\"] = (\n", " user_stats[\"total_likes\"] / user_stats[\"total_interactions\"]\n", " )\n", "\n", " taste_df = taste_df.merge(user_stats, on=\"user_id\", how=\"left\")\n", "\n", " # Filter by minimum interactions\n", " taste_df = taste_df[taste_df[\"total_interactions\"] >= min_interactions]\n", "\n", " # Normalize PageRank scores\n", " min_pr = taste_df[\"pagerank_score\"].min()\n", " max_pr = taste_df[\"pagerank_score\"].max()\n", " taste_df[\"taste_score\"] = (taste_df[\"pagerank_score\"] - min_pr) / (max_pr - min_pr)\n", "\n", " # Sort by taste score\n", " taste_df = taste_df.sort_values(\"taste_score\", ascending=False)\n", " taste_df[\"rank\"] = range(1, len(taste_df) + 1)\n", "\n", " return taste_df\n", "\n", "\n", "# Extract user taste scores\n", "user_taste_scores = extract_user_taste_scores(pagerank_scores, df_filtered)\n", "\n", "print(f\"Total users with taste scores: {len(user_taste_scores)}\")\n", "\n", "# Debug: Check why high like-rate users are appearing\n", "print(\"\\nDEBUG: Investigating high like-rate users in results...\")\n", "high_like_users = user_taste_scores[user_taste_scores[\"like_rate\"] > 0.9]\n", "print(f\"Found {len(high_like_users)} users with like_rate > 0.9 in the results\")\n", "\n", "if len(high_like_users) > 0:\n", " print(\"\\nChecking top 3 high like-rate users:\")\n", " for idx, (_, user) in enumerate(high_like_users.head(3).iterrows()):\n", " uid = user[\"user_id\"]\n", " print(f\"\\nUser {uid}:\")\n", " print(f\" - In bot_users set: {uid in bot_users}\")\n", " print(\n", " f\" - Stats in df_filtered: {user['total_interactions']} interactions, {user['total_likes']} likes\"\n", " )\n", "\n", " # Check their original stats\n", " orig_data = reactions_on_others[reactions_on_others[\"user_id\"] == uid]\n", " print(\n", " f\" - Stats in original data: {len(orig_data)} interactions, {(orig_data['reaction_type'] == 'L').sum()} likes\"\n", " )\n", "\n", " # Check what interactions remain in df_filtered\n", " filtered_data = df_filtered[df_filtered[\"user_id\"] == uid]\n", " print(f\" - Interactions in df_filtered: {len(filtered_data)}\")\n", " if len(filtered_data) > 0:\n", " print(\n", " f\" - Reaction types: {filtered_data['reaction_type'].value_counts().to_dict()}\"\n", " )\n", "print(f\"\\nTop 20 users with best taste:\")\n", "print(\n", " user_taste_scores.head(20)[\n", " [\n", " \"user_id\",\n", " \"taste_score\",\n", " \"total_interactions\",\n", " \"total_likes\",\n", " \"like_rate\",\n", " \"avg_play_count\",\n", " \"unique_clips\",\n", " ]\n", " ]\n", ")" ] }, { "cell_type": "code", "execution_count": 14, "id": "8c28d47f", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Checking specific users' taste scores...\n", "\n", "User 4619874 Analysis:\n", " - PageRank score: 0.000169\n", " - Selectivity weight: 2.90\n", " - Total likes: 99\n", "\n", " Clips this user liked:\n", " - Clip ccbada7e-ec0...: score=0.000005\n", " - Clip 2e0289e6-329...: score=0.000003\n", " - Clip 0f623ce3-5db...: score=0.000003\n", " - Clip 7678dcf5-aee...: score=0.000003\n", " - Clip b482b47d-c88...: score=0.000003\n", "\n", " Collaborative connections:\n", " Users who like the same clips:\n", " - User 51232507: weight=7.15\n", " - User 101217203: weight=3.92\n", " - User 65842865: weight=1.72\n", " - User 34630406: weight=2.69\n", " - User 74688582: weight=2.66\n" ] } ], "source": [ "# Check specific user's taste score (for debugging)\n", "def check_user_taste(\n", " user_id: int, pagerank_scores: dict, df: pd.DataFrame, user_weights: pd.DataFrame\n", "):\n", " \"\"\"Check a specific user's taste score and understand why\"\"\"\n", " user_node = f\"u_{user_id}\"\n", "\n", " if user_node not in pagerank_scores:\n", " print(f\"User {user_id} not found in PageRank scores (might not have any likes)\")\n", " return\n", "\n", " score = pagerank_scores[user_node]\n", " print(f\"\\nUser {user_id} Analysis:\")\n", " print(f\" - PageRank score: {score:.6f}\")\n", "\n", " # Get user's weight\n", " if user_id in user_weights.index:\n", " weight = user_weights.loc[user_id, \"selectivity_weight\"]\n", " print(f\" - Selectivity weight: {weight:.2f}\")\n", " else:\n", " print(f\" - Selectivity weight: Not found (default 1.0)\")\n", "\n", " # Get user's likes\n", " user_likes = df[(df[\"user_id\"] == user_id) & (df[\"reaction_type\"] == \"L\")]\n", " print(f\" - Total likes: {len(user_likes)}\")\n", "\n", " # Check what clips they liked and those clips' scores\n", " if len(user_likes) > 0:\n", " print(f\"\\n Clips this user liked:\")\n", " for _, like in user_likes.head(5).iterrows():\n", " clip_id = like[\"clip_id\"]\n", " clip_node = f\"c_{clip_id}\"\n", " if clip_node in pagerank_scores:\n", " clip_score = pagerank_scores[clip_node]\n", " print(f\" - Clip {clip_id[:12]}...: score={clip_score:.6f}\")\n", " else:\n", " print(f\" - Clip {clip_id[:12]}...: not in graph\")\n", "\n", " # Check who else likes the same clips (collaborative aspect)\n", " print(f\"\\n Collaborative connections:\")\n", " shared_taste_users = set()\n", " for _, like in user_likes.iterrows():\n", " clip_id = like[\"clip_id\"]\n", " # Find other users who liked this clip\n", " other_likers = df[\n", " (df[\"clip_id\"] == clip_id)\n", " & (df[\"reaction_type\"] == \"L\")\n", " & (df[\"user_id\"] != user_id)\n", " ]\n", " for other_uid in other_likers[\"user_id\"].unique()[:3]: # Just show first 3\n", " if other_uid in user_weights.index:\n", " other_weight = user_weights.loc[other_uid, \"selectivity_weight\"]\n", " shared_taste_users.add((other_uid, other_weight))\n", "\n", " if shared_taste_users:\n", " print(f\" Users who like the same clips:\")\n", " for other_uid, other_weight in list(shared_taste_users)[:5]:\n", " print(f\" - User {other_uid}: weight={other_weight:.2f}\")\n", "\n", "\n", "# Example: Check specific users\n", "print(\"Checking specific users' taste scores...\")\n", "# You can add specific user IDs here to debug\n", "example_users = [4619874] # Add the user ID mentioned\n", "for uid in example_users:\n", " check_user_taste(uid, pagerank_scores, df_filtered, user_weights)" ] }, { "cell_type": "code", "execution_count": 15, "id": "bd2f6a5e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "=== VALIDATION: Bot Detection Effectiveness ===\n", "✅ Bot detection successful: No users with like_rate > 80% in results\n", "\n", "=== Like Rate Distribution ===\n", "Overall like rate in data: 14.3%\n", "\n", "Top 100 users:\n", " - Mean like rate: 42.8%\n", " - Median like rate: 45.2%\n", " - Max like rate: 59.9%\n", " - Min like rate: 3.9%\n", "\n", "=== Selectivity Weights of Top Users ===\n", "User ID | Like Rate | Selectivity Weight | Taste Score\n", "------------------------------------------------------------\n", "78364115.0 | 59.9% | 1.7 | 1.000\n", "89271156.0 | 51.9% | 1.9 | 0.940\n", "75567253.0 | 29.5% | 3.4 | 0.911\n", "62795185.0 | 53.4% | 1.9 | 0.880\n", "72663343.0 | 50.5% | 2.0 | 0.851\n", "1173818.0 | 30.0% | 3.3 | 0.635\n", "88712285.0 | 43.9% | 2.3 | 0.527\n", "96883241.0 | 21.1% | 4.7 | 0.523\n", "5515095.0 | 43.5% | 2.3 | 0.492\n", "28605241.0 | 52.3% | 1.9 | 0.441\n", "86854140.0 | 31.1% | 3.2 | 0.436\n", "65956664.0 | 51.1% | 2.0 | 0.421\n", "6012821.0 | 47.1% | 2.1 | 0.419\n", "100990846.0 | 55.4% | 1.8 | 0.416\n", "95915360.0 | 55.1% | 1.8 | 0.416\n", "86499376.0 | 53.2% | 1.9 | 0.388\n", "85309214.0 | 28.1% | 3.6 | 0.375\n", "65961470.0 | 54.6% | 1.8 | 0.373\n", "63392178.0 | 44.9% | 2.2 | 0.371\n", "76055629.0 | 58.3% | 1.7 | 0.369\n", "\n", "============================================================\n" ] } ], "source": [ "# Validation: Ensure bot detection worked properly\n", "print(\"\\n=== VALIDATION: Bot Detection Effectiveness ===\")\n", "\n", "# Check if any high like-rate users made it through\n", "high_like_rate_threshold = 0.8\n", "high_like_rate_users = user_taste_scores[\n", " user_taste_scores[\"like_rate\"] > high_like_rate_threshold\n", "]\n", "\n", "if len(high_like_rate_users) > 0:\n", " print(\n", " f\"⚠️ WARNING: Found {len(high_like_rate_users)} users with like_rate > {high_like_rate_threshold:.0%}!\"\n", " )\n", " print(\"These users should have been caught by bot detection:\")\n", " print(\"\\nTop 10 high like-rate users that slipped through:\")\n", " print(\n", " high_like_rate_users.head(10)[\n", " [\n", " \"user_id\",\n", " \"like_rate\",\n", " \"total_interactions\",\n", " \"total_likes\",\n", " \"taste_score\",\n", " \"rank\",\n", " ]\n", " ]\n", " )\n", "\n", " # Show examples of what they liked\n", " print(\"\\nExamining what these users liked...\")\n", " example_user = high_like_rate_users.iloc[0][\"user_id\"]\n", " user_likes = df_filtered[\n", " (df_filtered[\"user_id\"] == example_user) & (df_filtered[\"reaction_type\"] == \"L\")\n", " ]\n", " print(f\"User {example_user} liked {len(user_likes)} clips\")\n", "\n", "else:\n", " print(\n", " f\"✅ Bot detection successful: No users with like_rate > {high_like_rate_threshold:.0%} in results\"\n", " )\n", "\n", "# Additional validation checks\n", "print(f\"\\n=== Like Rate Distribution ===\")\n", "print(\n", " f\"Overall like rate in data: {(df_filtered['reaction_type'] == 'L').sum() / len(df_filtered):.1%}\"\n", ")\n", "print(f\"\\nTop 100 users:\")\n", "top_100 = user_taste_scores.head(100)\n", "print(f\" - Mean like rate: {top_100['like_rate'].mean():.1%}\")\n", "print(f\" - Median like rate: {top_100['like_rate'].median():.1%}\")\n", "print(f\" - Max like rate: {top_100['like_rate'].max():.1%}\")\n", "print(f\" - Min like rate: {top_100['like_rate'].min():.1%}\")\n", "\n", "# Check selectivity distribution\n", "print(f\"\\n=== Selectivity Weights of Top Users ===\")\n", "top_20 = user_taste_scores.head(20)\n", "print(\"User ID | Like Rate | Selectivity Weight | Taste Score\")\n", "print(\"-\" * 60)\n", "for _, user in top_20.iterrows():\n", " uid = user[\"user_id\"]\n", " if uid in user_weights.index:\n", " weight = user_weights.loc[uid, \"selectivity_weight\"]\n", " print(\n", " f\"{uid} | {user['like_rate']:8.1%} | {weight:17.1f} | {user['taste_score']:.3f}\"\n", " )\n", " else:\n", " print(\n", " f\"{uid} | {user['like_rate']:8.1%} | {'N/A':>17} | {user['taste_score']:.3f}\"\n", " )\n", "\n", "print(\"\\n\" + \"=\" * 60)" ] }, { "cell_type": "code", "execution_count": 16, "id": "fc86a807", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Investigating potential bot-like behavior in top users:\n", "Users with taste_score > 0.8 and like_rate > 0.3:\n", " user_id taste_score like_rate total_interactions total_likes\n", "228 78364115 1.000000 0.598754 6101 3653\n", "613 89271156 0.940431 0.518962 6513 3380\n", "287 62795185 0.879819 0.533914 5278 2818\n", "344 72663343 0.850619 0.505412 4804 2428\n", "\n", "WARNING: Found 4 users with high taste scores but also high like rates!\n", "These users might be sophisticated bots that like popular content.\n", "\n", "Like rate statistics for top 100 users:\n", " - Mean like rate: 42.79%\n", " - Max like rate: 59.88%\n", " - Users with like_rate > 0.5: 36\n" ] } ], "source": [ "# Check if high like-rate users are still getting high taste scores\n", "print(\"\\nInvestigating potential bot-like behavior in top users:\")\n", "print(\"Users with taste_score > 0.8 and like_rate > 0.3:\")\n", "suspicious_users = user_taste_scores[\n", " (user_taste_scores[\"taste_score\"] > 0.8) & (user_taste_scores[\"like_rate\"] > 0.3)\n", "]\n", "print(\n", " suspicious_users[\n", " [\"user_id\", \"taste_score\", \"like_rate\", \"total_interactions\", \"total_likes\"]\n", " ]\n", ")\n", "\n", "if len(suspicious_users) > 0:\n", " print(\n", " f\"\\nWARNING: Found {len(suspicious_users)} users with high taste scores but also high like rates!\"\n", " )\n", " print(\"These users might be sophisticated bots that like popular content.\")\n", "\n", "# Also check the like rate distribution among top users\n", "print(f\"\\nLike rate statistics for top 100 users:\")\n", "top_100 = user_taste_scores.head(100)\n", "print(f\" - Mean like rate: {top_100['like_rate'].mean():.2%}\")\n", "print(f\" - Max like rate: {top_100['like_rate'].max():.2%}\")\n", "print(f\" - Users with like_rate > 0.5: {len(top_100[top_100['like_rate'] > 0.5])}\")" ] }, { "cell_type": "code", "execution_count": 17, "id": "be4f6c84", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "Top 20 users with best taste:\n", " user_id taste_score total_interactions total_likes like_rate \\\n", "228 78364115 1.000000 6101 3653 0.598754 \n", "613 89271156 0.940431 6513 3380 0.518962 \n", "692 75567253 0.910769 7209 2128 0.295187 \n", "287 62795185 0.879819 5278 2818 0.533914 \n", "344 72663343 0.850619 4804 2428 0.505412 \n", "522 1173818 0.635328 4832 1451 0.300290 \n", "304 88712285 0.526536 3062 1343 0.438602 \n", "240 96883241 0.523078 4014 848 0.211261 \n", "668 5515095 0.491851 3300 1436 0.435152 \n", "93 28605241 0.441019 3308 1729 0.522672 \n", "1043 86854140 0.435734 3571 1109 0.310557 \n", "411 65956664 0.420605 2997 1532 0.511178 \n", "1648 6012821 0.419117 2956 1393 0.471245 \n", "2 100990846 0.416339 2449 1356 0.553695 \n", "201 95915360 0.415847 2877 1584 0.550574 \n", "250 86499376 0.387642 2278 1211 0.531607 \n", "703 85309214 0.375416 3069 861 0.280547 \n", "666 65961470 0.373401 2158 1179 0.546339 \n", "33 63392178 0.371392 3232 1452 0.449257 \n", "1013 76055629 0.369239 1847 1077 0.583108 \n", "\n", " avg_play_count \n", "228 1.165711 \n", "613 1.445263 \n", "692 1.287696 \n", "287 1.709170 \n", "344 1.201499 \n", "522 1.152318 \n", "304 1.183867 \n", "240 1.059791 \n", "668 1.115152 \n", "93 1.174123 \n", "1043 1.385046 \n", "411 1.275609 \n", "1648 1.502368 \n", "2 1.265006 \n", "201 1.278415 \n", "250 1.165496 \n", "703 1.327794 \n", "666 1.132530 \n", "33 1.626238 \n", "1013 1.078506 \n" ] } ], "source": [ "# No second-pass filtering needed - bot detection handled in first pass\n", "print(\"\\nTop 20 users with best taste:\")\n", "print(\n", " user_taste_scores.head(20)[\n", " [\n", " \"user_id\",\n", " \"taste_score\",\n", " \"total_interactions\",\n", " \"total_likes\",\n", " \"like_rate\",\n", " \"avg_play_count\",\n", " ]\n", " ]\n", ")" ] }, { "cell_type": "code", "execution_count": 18, "id": "d276aa71", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "=== MR TREE (User 4619874) ANALYSIS ===\n", "✅ Mr tree was NOT flagged as a bot\n", "\n", "Original data stats:\n", " - Total interactions: 287\n", " - Likes: 99\n", " - Like rate: 34.5%\n", "\n", "Filtered data stats:\n", " - Total interactions: 287\n", " - Likes: 99\n", "\n", "✅ Mr tree IS in final taste scores!\n", " - Taste score: 0.0556\n", " - Rank: 322.0 out of 22959\n", " - PageRank score: 0.000169\n", " - Like rate: 34.5%\n", "\n", "Selectivity weight: 2.90\n" ] } ], "source": [ "# Check Mr tree's status\n", "mr_tree_id = 4619874\n", "print(f\"\\n=== MR TREE (User {mr_tree_id}) ANALYSIS ===\")\n", "\n", "# Check if in bot_users\n", "if mr_tree_id in bot_users:\n", " print(f\"❌ Mr tree was incorrectly flagged as a bot!\")\n", "else:\n", " print(f\"✅ Mr tree was NOT flagged as a bot\")\n", "\n", "# Check stats in original data\n", "mr_tree_orig = reactions_on_others[reactions_on_others[\"user_id\"] == mr_tree_id]\n", "print(f\"\\nOriginal data stats:\")\n", "print(f\" - Total interactions: {len(mr_tree_orig)}\")\n", "print(f\" - Likes: {(mr_tree_orig['reaction_type'] == 'L').sum()}\")\n", "print(\n", " f\" - Like rate: {(mr_tree_orig['reaction_type'] == 'L').sum() / len(mr_tree_orig):.1%}\"\n", ")\n", "\n", "# Check if in filtered data\n", "mr_tree_filtered = df_filtered[df_filtered[\"user_id\"] == mr_tree_id]\n", "print(f\"\\nFiltered data stats:\")\n", "print(f\" - Total interactions: {len(mr_tree_filtered)}\")\n", "print(f\" - Likes: {(mr_tree_filtered['reaction_type'] == 'L').sum()}\")\n", "\n", "# Check if in user_taste_scores\n", "mr_tree_taste = user_taste_scores[user_taste_scores[\"user_id\"] == mr_tree_id]\n", "if len(mr_tree_taste) > 0:\n", " print(f\"\\n✅ Mr tree IS in final taste scores!\")\n", " print(f\" - Taste score: {mr_tree_taste.iloc[0]['taste_score']:.4f}\")\n", " print(f\" - Rank: {mr_tree_taste.iloc[0]['rank']} out of {len(user_taste_scores)}\")\n", " print(f\" - PageRank score: {mr_tree_taste.iloc[0]['pagerank_score']:.6f}\")\n", " print(f\" - Like rate: {mr_tree_taste.iloc[0]['like_rate']:.1%}\")\n", "else:\n", " print(f\"\\n❌ Mr tree is NOT in final taste scores!\")\n", "\n", "# Check selectivity weight\n", "if mr_tree_id in user_weights.index:\n", " print(\n", " f\"\\nSelectivity weight: {user_weights.loc[mr_tree_id, 'selectivity_weight']:.2f}\"\n", " )" ] }, { "cell_type": "code", "execution_count": 19, "id": "45d9747e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "=== VALIDATION: Selective vs Indiscriminate Users ===\n", "\n", "Top 100 users:\n", " - Mean like rate: 42.8%\n", " - Median like rate: 45.2%\n", " - Users with like_rate < 20%: 6\n", " - Users with like_rate < 40%: 35\n", " - Users with like_rate > 60%: 0\n", "\n", "Bottom 100 users:\n", " - Mean like rate: 49.8%\n", " - Median like rate: 50.0%\n", "\n", "✅ SUCCESS: No users with >60% like rate in top 100!\n", "\n", "Correlation between like_rate and taste_score: 0.152\n", "❌ Problem: Positive correlation means indiscriminate users get higher scores\n" ] } ], "source": [ "# Validate that selective users get higher scores\n", "print(\"\\n=== VALIDATION: Selective vs Indiscriminate Users ===\")\n", "\n", "# Check like rate distribution in top users\n", "top_100 = user_taste_scores.head(100)\n", "bottom_100 = user_taste_scores.tail(100)\n", "\n", "print(f\"\\nTop 100 users:\")\n", "print(f\" - Mean like rate: {top_100['like_rate'].mean():.1%}\")\n", "print(f\" - Median like rate: {top_100['like_rate'].median():.1%}\")\n", "print(f\" - Users with like_rate < 20%: {len(top_100[top_100['like_rate'] < 0.2])}\")\n", "print(f\" - Users with like_rate < 40%: {len(top_100[top_100['like_rate'] < 0.4])}\")\n", "print(f\" - Users with like_rate > 60%: {len(top_100[top_100['like_rate'] > 0.6])}\")\n", "\n", "print(f\"\\nBottom 100 users:\")\n", "print(f\" - Mean like rate: {bottom_100['like_rate'].mean():.1%}\")\n", "print(f\" - Median like rate: {bottom_100['like_rate'].median():.1%}\")\n", "\n", "# Check if any high like-rate users are in top 100\n", "high_like_in_top = top_100[top_100[\"like_rate\"] > 0.6]\n", "if len(high_like_in_top) > 0:\n", " print(\n", " f\"\\n⚠️ WARNING: Found {len(high_like_in_top)} users with >60% like rate in top 100!\"\n", " )\n", " print(\"These users might still be gaming the system:\")\n", " print(\n", " high_like_in_top[\n", " [\"user_id\", \"like_rate\", \"total_interactions\", \"taste_score\"]\n", " ].head()\n", " )\n", "else:\n", " print(f\"\\n✅ SUCCESS: No users with >60% like rate in top 100!\")\n", "\n", "# Check correlation between selectivity and taste score\n", "if len(user_taste_scores) > 10:\n", " correlation = user_taste_scores[\"like_rate\"].corr(user_taste_scores[\"taste_score\"])\n", " print(f\"\\nCorrelation between like_rate and taste_score: {correlation:.3f}\")\n", " if correlation < 0:\n", " print(\"✅ Good: Negative correlation means selective users get higher scores\")\n", " else:\n", " print(\n", " \"❌ Problem: Positive correlation means indiscriminate users get higher scores\"\n", " )" ] }, { "cell_type": "code", "execution_count": 20, "id": "9b8c028e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Validation Results:\n", "Average quality of clips liked by top 100 taste users: 0.005\n", "Average quality of clips liked by bottom 100 taste users: 0.261\n", "Quality difference: -0.256\n", "\n", "Average like rate of top 100 users: 42.79%\n", "Average like rate of bottom 100 users: 49.83%\n", "\n", "Overlap in liked clips between top and bottom users: 0.1%\n" ] } ], "source": [ "# Step 6: Validation - Check if high-taste users like similar quality clips\n", "def validate_taste_scores(\n", " user_taste_scores: pd.DataFrame,\n", " df: pd.DataFrame,\n", " clip_quality: pd.DataFrame,\n", " top_n: int = 100,\n", ") -> None:\n", " \"\"\"Validate that high-taste users tend to like high-quality clips.\"\"\"\n", "\n", " # Get top taste users\n", " top_users = user_taste_scores.head(top_n)[\"user_id\"].tolist()\n", " bottom_users = user_taste_scores.tail(top_n)[\"user_id\"].tolist()\n", "\n", " # Get clips liked by each group\n", " top_user_likes = df[(df[\"user_id\"].isin(top_users)) & (df[\"reaction_type\"] == \"L\")][\n", " \"clip_id\"\n", " ].tolist()\n", " bottom_user_likes = df[\n", " (df[\"user_id\"].isin(bottom_users)) & (df[\"reaction_type\"] == \"L\")\n", " ][\"clip_id\"].tolist()\n", "\n", " # Calculate average quality of liked clips\n", " top_clips_quality = clip_quality.loc[\n", " clip_quality.index.isin(top_user_likes), \"quality_score_norm\"\n", " ].mean()\n", "\n", " bottom_clips_quality = clip_quality.loc[\n", " clip_quality.index.isin(bottom_user_likes), \"quality_score_norm\"\n", " ].mean()\n", "\n", " print(f\"Validation Results:\")\n", " print(\n", " f\"Average quality of clips liked by top {top_n} taste users: {top_clips_quality:.3f}\"\n", " )\n", " print(\n", " f\"Average quality of clips liked by bottom {top_n} taste users: {bottom_clips_quality:.3f}\"\n", " )\n", " print(f\"Quality difference: {top_clips_quality - bottom_clips_quality:.3f}\")\n", "\n", " # Also show like rates\n", " top_users_df = user_taste_scores[user_taste_scores[\"user_id\"].isin(top_users)]\n", " bottom_users_df = user_taste_scores[user_taste_scores[\"user_id\"].isin(bottom_users)]\n", " print(\n", " f\"\\nAverage like rate of top {top_n} users: {top_users_df['like_rate'].mean():.2%}\"\n", " )\n", " print(\n", " f\"Average like rate of bottom {top_n} users: {bottom_users_df['like_rate'].mean():.2%}\"\n", " )\n", "\n", " # Check overlap in liked clips\n", " top_clips_set = set(top_user_likes)\n", " bottom_clips_set = set(bottom_user_likes)\n", " overlap = len(top_clips_set & bottom_clips_set) / len(\n", " top_clips_set | bottom_clips_set\n", " )\n", " print(f\"\\nOverlap in liked clips between top and bottom users: {overlap:.1%}\")\n", "\n", "\n", "validate_taste_scores(user_taste_scores, df_filtered, clip_quality)" ] }, { "cell_type": "markdown", "id": "636d2ce8", "metadata": {}, "source": [ "## Key Issue Found\n", "\n", "The main problem is that users with 99%+ like rates are appearing in the top results. This happens because:\n", "\n", "1. **Bot detection threshold is too lenient**: Only flags users with >50% like rate AND >50 interactions\n", "2. **These users ARE in df_filtered**: They passed through bot detection somehow\n", "3. **PageRank rewards them**: Users who like everything get high scores if they like popular content\n", "\n", "Run the notebook with the debug code to see exactly what's happening with these suspicious users.\n" ] }, { "cell_type": "code", "execution_count": 21, "id": "1c56367e", "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Visualize taste score distribution\n", "fig, axes = plt.subplots(2, 2, figsize=(15, 10))\n", "\n", "# 1. Taste score distribution\n", "axes[0, 0].hist(user_taste_scores[\"taste_score\"], bins=50, edgecolor=\"black\", alpha=0.7)\n", "axes[0, 0].set_xlabel(\"Taste Score\")\n", "axes[0, 0].set_ylabel(\"Number of Users\")\n", "axes[0, 0].set_title(\"Distribution of User Taste Scores (PageRank)\")\n", "\n", "# 2. Taste score vs play count\n", "axes[0, 1].scatter(\n", " user_taste_scores[\"avg_play_count\"],\n", " user_taste_scores[\"taste_score\"],\n", " alpha=0.5,\n", " s=20,\n", ")\n", "axes[0, 1].set_xlabel(\"Average Play Count\")\n", "axes[0, 1].set_ylabel(\"Taste Score\")\n", "axes[0, 1].set_title(\"Taste Score vs Engagement\")\n", "\n", "# 3. Taste score vs number of interactions\n", "axes[1, 0].scatter(\n", " user_taste_scores[\"total_interactions\"],\n", " user_taste_scores[\"taste_score\"],\n", " alpha=0.5,\n", " s=20,\n", ")\n", "axes[1, 0].set_xlabel(\"Total Interactions\")\n", "axes[1, 0].set_ylabel(\"Taste Score\")\n", "axes[1, 0].set_xscale(\"log\")\n", "axes[1, 0].set_title(\"Taste Score vs Activity Level\")\n", "\n", "# 4. Compare bot vs non-bot behavior\n", "bot_interactions = reactions_on_others[reactions_on_others[\"user_id\"].isin(bot_users)]\n", "bot_stats = bot_interactions.groupby(\"user_id\")[\"play_count\"].mean().mean()\n", "normal_stats = df_filtered.groupby(\"user_id\")[\"play_count\"].mean().mean()\n", "\n", "axes[1, 1].bar(\n", " [\"Bot Users\", \"Normal Users\"],\n", " [bot_stats, normal_stats],\n", " color=[\"red\", \"green\"],\n", " alpha=0.7,\n", ")\n", "axes[1, 1].set_ylabel(\"Average Play Count\")\n", "axes[1, 1].set_title(\"Bot vs Normal User Consumption Behavior\")\n", "\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 22, "id": "fb00cd6b", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "=== PAGERANK TASTE ANALYSIS SUMMARY ===\n", "\n", "1. Bot Detection:\n", " - Detected 4277 bot users\n", " - Removed 1210799 bot interactions\n", " - Bot avg play count: 1.75 vs Normal: 1.17\n", "\n", "2. Selectivity Weight Distribution:\n", " - Total users with weights: 149695\n", " - Mean selectivity weight: 26.26\n", " - Median selectivity weight: 20.00\n", " - Users with weight > 10 (very selective): 92416\n", " - Users with weight > 20 (extremely selective): 67465\n", "\n", " Weight percentiles:\n", " - 10th percentile: 8.18\n", " - 25th percentile: 10.00\n", " - 50th percentile: 20.00\n", " - 75th percentile: 50.00\n", " - 90th percentile: 50.00\n", " - 95th percentile: 50.00\n", " - 99th percentile: 50.00\n", "\n", "3. User Taste Distribution:\n", " - Total users analyzed: 22959\n", " - Average taste score: 0.006\n", " - Std deviation: 0.024\n", "\n", "4. Curator Candidates (taste >= 0.8, interactions >= 20):\n", " - Found 5 potential curators\n", " - They have 1.4x average play count\n", " - They interact with 5981 unique clips on average\n" ] } ], "source": [ "# Summary and Export Results\n", "print(\"=== PAGERANK TASTE ANALYSIS SUMMARY ===\\n\")\n", "\n", "# Key statistics\n", "print(f\"1. Bot Detection:\")\n", "print(f\" - Detected {len(bot_users)} bot users\")\n", "print(f\" - Removed {len(reactions_on_others) - len(df_filtered)} bot interactions\")\n", "print(f\" - Bot avg play count: {bot_stats:.2f} vs Normal: {normal_stats:.2f}\")\n", "\n", "print(f\"\\n2. Selectivity Weight Distribution:\")\n", "print(f\" - Total users with weights: {len(user_weights)}\")\n", "print(f\" - Mean selectivity weight: {user_weights['selectivity_weight'].mean():.2f}\")\n", "print(\n", " f\" - Median selectivity weight: {user_weights['selectivity_weight'].median():.2f}\"\n", ")\n", "print(\n", " f\" - Users with weight > 10 (very selective): {len(user_weights[user_weights['selectivity_weight'] > 10])}\"\n", ")\n", "print(\n", " f\" - Users with weight > 20 (extremely selective): {len(user_weights[user_weights['selectivity_weight'] > 20])}\"\n", ")\n", "\n", "# Show weight distribution by percentiles\n", "percentiles = [10, 25, 50, 75, 90, 95, 99]\n", "print(f\"\\n Weight percentiles:\")\n", "for p in percentiles:\n", " value = user_weights[\"selectivity_weight\"].quantile(p / 100)\n", " print(f\" - {p}th percentile: {value:.2f}\")\n", "\n", "print(f\"\\n3. User Taste Distribution:\")\n", "print(f\" - Total users analyzed: {len(user_taste_scores)}\")\n", "print(f\" - Average taste score: {user_taste_scores['taste_score'].mean():.3f}\")\n", "print(f\" - Std deviation: {user_taste_scores['taste_score'].std():.3f}\")\n", "\n", "# Find curator candidates\n", "high_taste_active = user_taste_scores[\n", " (user_taste_scores[\"taste_score\"] >= 0.8)\n", " & (user_taste_scores[\"total_interactions\"] >= 20)\n", "]\n", "\n", "print(f\"\\n4. Curator Candidates (taste >= 0.8, interactions >= 20):\")\n", "print(f\" - Found {len(high_taste_active)} potential curators\")\n", "print(\n", " f\" - They have {high_taste_active['avg_play_count'].mean():.1f}x average play count\"\n", ")\n", "print(\n", " f\" - They interact with {high_taste_active['unique_clips'].mean():.0f} unique clips on average\"\n", ")\n", "\n", "# # Save results\n", "# user_taste_scores.to_csv(\n", "# \"/home/tony/Data/Preference/pagerank_user_taste_scores.csv\", index=False\n", "# )\n", "# clip_quality.to_csv(\"/home/tony/Data/Preference/pagerank_clip_quality.csv\")\n", "# high_taste_active.to_csv(\n", "# \"/home/tony/Data/Preference/curator_candidates.csv\", index=False\n", "# )\n", "\n", "# print(f\"\\n4. Files saved:\")\n", "# print(\n", "# f\" - User taste scores: /home/tony/Data/Preference/pagerank_user_taste_scores.csv\"\n", "# )\n", "# print(f\" - Clip quality scores: /home/tony/Data/Preference/pagerank_clip_quality.csv\")\n", "# print(f\" - Curator candidates: /home/tony/Data/Preference/curator_candidates.csv\")" ] }, { "cell_type": "markdown", "id": "055f049f", "metadata": {}, "source": [ "## Summary of Key Issues Found\n", "\n", "### 1. Clip Quality Calculation was Flawed\n", "- **Problem**: Top \"quality\" clips had only 1 like but high play counts\n", "- **Fix**: Now requires minimum 3 unique likers and weights unique likers heavily (40%)\n", "- **Result**: Only clips liked by multiple users can be considered \"good\"\n", "\n", "### 2. Bot Detection was Too Aggressive \n", "- **Problem**: Removed 50% of all interactions, leaving too few users\n", "- **Fix**: Adjusted thresholds - now only flag users with >50% like rate AND 50+ interactions\n", "- **Result**: More reasonable filtering that preserves genuine selective users\n", "\n", "### 3. Neutral Listens are Critical\n", "- **Finding**: 60.6% of interactions are neutral listens (no like/dislike)\n", "- **Insight**: Good taste users listen to many clips but only like the exceptional ones\n", "- **Implementation**: All interactions now contribute to the PageRank graph\n", "\n", "### 4. Data Sparsity Issue\n", "- Most clips have very few likes after bot filtering\n", "- This makes it hard to identify consensus about quality\n", "- Need either more data or different approach for sparse interaction data\n" ] }, { "cell_type": "markdown", "id": "6fae43e4", "metadata": {}, "source": [ "\n" ] }, { "cell_type": "markdown", "id": "ee982372", "metadata": {}, "source": [ "## Key Insights\n", "\n", "### Including Neutral Listens Changes Everything\n", "\n", "By counting all interactions (not just likes/dislikes), we reveal the true behavior patterns:\n", "\n", "1. **Real Users are Selective**: They listen to many clips but only like the truly good ones (low like rate)\n", "2. **Bots are Indiscriminate**: They like most of what they interact with (high like rate) and rarely just listen without reacting\n", "3. **Engagement Matters**: Neutral listens with high play counts show genuine interest even without explicit likes\n", "\n", "### PageRank Captures Good Taste\n", "\n", "The algorithm naturally identifies users with good taste through recursive scoring:\n", "\n", "1. **Selective Users Win**: Users who listen to many but like few get higher scores when those few likes align with other high-score users\n", "2. **Bot Filtering is Critical**: By removing users with high like rates and low consumption, we prevent bots from polluting the taste network\n", "3. **Quality Emerges from Consensus**: Clips become \"good\" when selectively liked by users who themselves have good taste\n", "\n", "This creates a virtuous circle where discerning users and quality content mutually identify each other.\n" ] }, { "cell_type": "markdown", "id": "057666c4", "metadata": {}, "source": [ "## Improvements Implemented\n", "\n", "This notebook has been updated with the following improvements based on the instructions:\n", "\n", "### 1. **Bot Detection (Fixed)**\n", "- ✅ Threshold updated to >80% like rate (as per instructions)\n", "- ✅ Added detection for users with >100 likes but avg_play_per_like < 0.1\n", "- ✅ Consolidated redundant criteria for cleaner code\n", "- ✅ Removed second-pass filtering - all bots caught in first pass\n", "\n", "### 2. **Selective User Vote Weighting (New Feature)**\n", "- ✅ Implemented weight = total_interactions / total_likes\n", "- ✅ Selective users (few likes) get higher weights up to 50\n", "- ✅ Example: 100 interactions, 4 likes → weight = 25\n", "- ✅ Example: 100 interactions, 50 likes → weight = 2\n", "\n", "### 3. **Graph Construction (Enhanced)**\n", "- ✅ User → Clip edges weighted by: selectivity_weight × (1 + log(play_count))\n", "- ✅ Selective users who actually listen have the most influence\n", "- ✅ Neutral listens included with lower weight (0.1 + 0.1 × log(play_count))\n", "- ✅ Dislikes weighted negatively by selectivity\n", "\n", "### 4. **Validation & Monitoring (Added)**\n", "- ✅ Automatic check for users with >80% like rate in results\n", "- ✅ Selectivity weight distribution statistics\n", "- ✅ Clear warnings if bot detection fails\n", "- ✅ Top user selectivity weights displayed\n", "\n", "### 5. **Code Quality**\n", "- ✅ Removed duplicate bot detection criteria\n", "- ✅ Cleaned up commented debug code\n", "- ✅ Added comprehensive documentation\n", "- ✅ Better organized summary statistics\n", "\n", "### Key Insight\n", "The algorithm now properly rewards **selective users** - those who listen to many clips but only like the truly exceptional ones. Their votes carry more weight in determining clip quality and user taste scores through the PageRank algorithm.\n" ] }, { "cell_type": "code", "execution_count": 23, "id": "e3fce4f7", "metadata": {}, "outputs": [], "source": [ "# 4619874 is Mr tree" ] }, { "cell_type": "markdown", "id": "f94f98eb", "metadata": {}, "source": [ "## How the Collaborative Filtering Works\n", "\n", "The simplified algorithm works as follows:\n", "\n", "1. **User Selectivity Weights**: \n", " - Users who rarely like things get high weights (up to 50)\n", " - Users who like everything get low weights (~2)\n", "\n", "2. **Bipartite Graph**:\n", " - Only includes likes (positive signals)\n", " - User → Clip edges weighted by user's selectivity\n", " - Clip → User edges have weight 1.0\n", "\n", "3. **PageRank Flow**:\n", " - Selective users (high weight) strongly boost clips they like\n", " - Users who like the same clips as selective users get higher scores\n", " - This creates collaborative filtering where taste propagates through shared preferences\n", "\n", "4. **Key Insight**:\n", " - If you like clips that selective users also like, you probably have good taste\n", " - If you like clips that only non-selective users like, you probably don't\n", "\n", "### Debugging Tips:\n", "- Use `check_user_taste()` to understand why a specific user has a certain score\n", "- Look at clip scores to see which clips are considered \"good\" by the algorithm\n", "- Check if users share taste with selective users for high scores\n", "\n" ] }, { "cell_type": "markdown", "id": "69becb5c", "metadata": {}, "source": [ "## Key Addition: Weighted Likes in Clip Quality\n", "\n", "The clip quality calculation now uses **weighted likes** instead of raw like counts. This is crucial because:\n", "\n", "1. **Selective users' likes matter more**: A like from someone who rarely likes anything (weight=25) contributes much more to clip quality than a like from someone who likes everything (weight=2).\n", "\n", "2. **Bot-resistant quality scores**: Even if bots slip through detection, their low selectivity weights minimize their impact on clip quality.\n", "\n", "3. **True quality emerges**: Clips that are liked by discerning, selective users will have higher quality scores than clips liked by indiscriminate users.\n", "\n", "### Example:\n", "- Clip A: 10 likes from users who like everything (avg weight = 2) → weighted likes = 20\n", "- Clip B: 5 likes from selective users (avg weight = 10) → weighted likes = 50\n", "- Result: Clip B has higher quality despite fewer raw likes\n", "\n", "This ensures the PageRank algorithm propagates quality signals from truly good clips, as determined by selective users.\n" ] }, { "cell_type": "code", "execution_count": 24, "id": "6e2d605a", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.00016855430293135392" ] }, "execution_count": 24, "metadata": {}, "output_type": "execute_result" } ], "source": [ "pagerank_scores[\"u_4619874\"]" ] }, { "cell_type": "code", "execution_count": 25, "id": "cf8f6e2b", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
user_idpagerank_scoretotal_interactionstotal_likesavg_play_countunique_clipslike_ratetaste_scorerank
108946198740.000169287991.3240422870.3449480.055622322
\n", "
" ], "text/plain": [ " user_id pagerank_score total_interactions total_likes \\\n", "1089 4619874 0.000169 287 99 \n", "\n", " avg_play_count unique_clips like_rate taste_score rank \n", "1089 1.324042 287 0.344948 0.055622 322 " ] }, "execution_count": 25, "metadata": {}, "output_type": "execute_result" } ], "source": [ "user_taste_scores[user_taste_scores[\"user_id\"] == 4619874]" ] }, { "cell_type": "code", "execution_count": 26, "id": "0b03fe26", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
clip_iduser_idupdated_atplay_countskip_countflaggedflagged_reasonreaction_typeis_pro_userclip_creator_id
30974920f542bb-bcca-4b1c-a033-8cf4e9e586ae46198742025-07-24 21:09:17.960218+00:0010FalseNoneNoneTrue6227557.0
366091a3915b43-e78d-4fb7-b025-d6e0e06e7c3046198742025-07-24 21:31:19.095337+00:0010FalseNoneNoneTrue6227557.0
3957090b3bd2a1-5d6d-45d8-a7d2-07120107257f46198742025-07-24 21:42:38.563529+00:0010FalseNoneNoneTrue6227557.0
4368009845cb95-f2fe-47fe-b7b7-3a3e89bdf64246198742025-07-24 22:03:54.986201+00:0020FalseNoneNoneTrue6227557.0
45067318597fc6-0b8c-431b-ad18-7067e217474446198742025-07-24 22:06:54.510620+00:0020FalseNoneNoneTrue6227557.0
.................................
126080208a628be62-4108-402c-abab-c91367751d9346198742025-08-18 10:21:19.333225+00:0040FalseNoneLTrue14688.0
126130329c642462a-5751-43d7-8e58-82b21d5be32646198742025-08-18 10:31:07.121207+00:0040FalseNoneLTrue63411148.0
126139764cea2fd30-ef97-478c-9b13-5858fd86758146198742025-08-15 13:00:21.868782+00:0040FalseNoneLTrue100835929.0
126146341fa50cde9-1d30-4d84-9900-33f7602180e146198742025-08-15 01:39:57.286158+00:0030FalseNoneLTrue174639.0
1266513778c159961-879e-4218-8175-c8a4ada1481946198742025-08-08 13:08:38.886776+00:0010FalseNoneNoneTrue105757528.0
\n", "

287 rows × 10 columns

\n", "
" ], "text/plain": [ " clip_id user_id \\\n", "309749 20f542bb-bcca-4b1c-a033-8cf4e9e586ae 4619874 \n", "366091 a3915b43-e78d-4fb7-b025-d6e0e06e7c30 4619874 \n", "395709 0b3bd2a1-5d6d-45d8-a7d2-07120107257f 4619874 \n", "436800 9845cb95-f2fe-47fe-b7b7-3a3e89bdf642 4619874 \n", "450673 18597fc6-0b8c-431b-ad18-7067e2174744 4619874 \n", "... ... ... \n", "126080208 a628be62-4108-402c-abab-c91367751d93 4619874 \n", "126130329 c642462a-5751-43d7-8e58-82b21d5be326 4619874 \n", "126139764 cea2fd30-ef97-478c-9b13-5858fd867581 4619874 \n", "126146341 fa50cde9-1d30-4d84-9900-33f7602180e1 4619874 \n", "126651377 8c159961-879e-4218-8175-c8a4ada14819 4619874 \n", "\n", " updated_at play_count skip_count flagged \\\n", "309749 2025-07-24 21:09:17.960218+00:00 1 0 False \n", "366091 2025-07-24 21:31:19.095337+00:00 1 0 False \n", "395709 2025-07-24 21:42:38.563529+00:00 1 0 False \n", "436800 2025-07-24 22:03:54.986201+00:00 2 0 False \n", "450673 2025-07-24 22:06:54.510620+00:00 2 0 False \n", "... ... ... ... ... \n", "126080208 2025-08-18 10:21:19.333225+00:00 4 0 False \n", "126130329 2025-08-18 10:31:07.121207+00:00 4 0 False \n", "126139764 2025-08-15 13:00:21.868782+00:00 4 0 False \n", "126146341 2025-08-15 01:39:57.286158+00:00 3 0 False \n", "126651377 2025-08-08 13:08:38.886776+00:00 1 0 False \n", "\n", " flagged_reason reaction_type is_pro_user clip_creator_id \n", "309749 None None True 6227557.0 \n", "366091 None None True 6227557.0 \n", "395709 None None True 6227557.0 \n", "436800 None None True 6227557.0 \n", "450673 None None True 6227557.0 \n", "... ... ... ... ... \n", "126080208 None L True 14688.0 \n", "126130329 None L True 63411148.0 \n", "126139764 None L True 100835929.0 \n", "126146341 None L True 174639.0 \n", "126651377 None None True 105757528.0 \n", "\n", "[287 rows x 10 columns]" ] }, "execution_count": 26, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df_filtered[df_filtered[\"user_id\"] == 4619874]" ] }, { "cell_type": "markdown", "id": "3c5bf4c8", "metadata": {}, "source": [ "## 🔧 FIXES APPLIED TO ADDRESS MR TREE'S LOW SCORE\n", "\n", "### Problems Found:\n", "1. **Like rate calculation was broken** - It only counted non-null reaction_type values, making users who liked everything appear to have 100% like rate\n", "2. **Bot detection too lenient** - Users with 60-80% like rates were passing through\n", "3. **PageRank rewarded wrong behavior** - Users who liked everything got highest scores\n", "4. **Second-pass filter too aggressive** - Removed almost all users, leaving only 164\n", "\n", "### Fixes Applied:\n", "1. ✅ **Fixed like rate calculation** - Now properly counts ALL interactions including neutral listens\n", "2. ✅ **Lowered bot threshold to 60%** - More aggressive filtering (mean like rate is only 5.6%)\n", "3. ✅ **Modified graph weights** - Clip→User edges now weighted by selectivity, so selective users benefit more\n", "4. ✅ **Removed second-pass filter** - All bot detection handled in first pass\n", "5. ✅ **Added Mr tree validation** - Specific check to ensure he appears with appropriate score\n", "\n", "### Expected Results:\n", "- Mr tree (34.5% like rate) should now appear in results with a reasonable score\n", "- Top users should be selective (low like rates), not indiscriminate likers\n", "- Bot users with >60% like rates should be filtered out\n", "- Many more users should appear in final results (not just 164)\n" ] }, { "cell_type": "markdown", "id": "fd011cac", "metadata": {}, "source": [ "\n" ] }, { "cell_type": "code", "execution_count": null, "id": "c486531a", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "suno_env_dev", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.15" } }, "nbformat": 4, "nbformat_minor": 5 }