Upload folder using huggingface_hub

README.md

@@ -47,321 +47,107 @@ dataset_info:
 
 ## Dataset Summary
 
-This dataset contains movie metadata, plot summaries, and semantic embeddings for feature-length films released between 1930 and 2024. The dataset was created for analyzing temporal semantic drift in movie plots using embedding-based techniques, as part of the research project "Plot Twists Over Time: How Movie Stories Have Changed Over 95 Years". It includes comprehensive movie metadata from Wikidata, TMDb, and Wikipedia, along with dense and sparse embeddings generated using the BAAI/bge-m3 model.
+This dataset contains movie metadata, plot summaries, and semantic embeddings for ~92,000 feature-length films (1930-2024). Created for temporal semantic drift analysis, it includes metadata from Wikidata, TMDb, and Wikipedia, along with dense (1024-dim) and sparse embeddings generated using BAAI/bge-m3.
 
-**Dataset Size**: ~92,000 movies with plot summaries and embeddings
-
-**Year Range**: 1930-2024
-
-**Primary Use Case**: Temporal semantic analysis, genre classification, concept extraction, and embedding-based movie similarity analysis
+**Key Statistics**: ~92,000 movies | Year range: 1930-2024 | Embedding dimension: 1024 | Average plot length: ~1,500 characters
 
 ## Dataset Structure
 
-The dataset consists of the following files:
+### Core Files
 
-### Core Data Files
+- **`final_dataset.csv`** (200 MB): Movie metadata with 39 columns (title, year, genre, plot, directors, actors, TMDb metrics, etc.)
+- **`final_dense_embeddings.npy`** (380 MB): Dense embeddings array `(N, 1024)` using BGE-M3 CLS token aggregation
+- **`final_dense_movie_ids.npy`** (7 MB): Wikidata QIDs corresponding to embeddings (index-aligned: `embeddings[i]` ↔ `movie_ids[i]`)
+- **`final_sparse_embeddings.npz`** (500 MB): Sparse lexical weights for concept extraction (token_indices, weights, movie_ids)
 
-- **`final_dataset.csv`**: Consolidated movie metadata CSV file containing all movie information
-  - Contains 39 columns with movie metadata (title, year, genre, plot, etc.)
-  - Includes movies from 1930-2024
-  - All movies have corresponding embeddings in the embedding files
+### Additional Files
 
-- **`final_dense_embeddings.npy`**: NumPy array of dense embeddings
-  - Shape: `(N, 1024)` where N is the number of movies
-  - Generated using BAAI/bge-m3 model with CLS token aggregation
-  - Each embedding corresponds to a movie plot summary
+- **`knn_faiss_novelty.csv`**: Novelty scores and neighbor information
+- **`umap_cluster_trajectories.png`**: UMAP visualization of embeddings
+- **`concept_space/`**: WordNet-based concept vocabulary and embeddings for semantic mapping
 
-- **`final_dense_movie_ids.npy`**: NumPy array of movie IDs (Wikidata QIDs)
-  - Shape: `(N,)`
-  - Contains Wikidata QIDs (e.g., "Q1931001") corresponding to each embedding
-  - Index alignment: `final_dense_embeddings[i]` corresponds to `final_dense_movie_ids[i]`
+## Data Collection Pipeline
 
-- **`final_sparse_embeddings.npz`**: Sparse lexical weights (optional)
-  - Contains token-level lexical weights from BGE-M3 model
-  - Format: NPZ file with keys:
-    - `token_indices`: List of arrays, one per movie, containing token IDs with non-zero weights
-    - `weights`: List of arrays, one per movie, containing corresponding lexical weights
-    - `movie_ids`: Array of movie IDs corresponding to the lexical weights
-  - Used for concept extraction and fine-grained semantic analysis
+Four-step pipeline: (1) **Wikidata**: Query SPARQL for movie metadata (1930-2024, ~8K/year), filter feature films; (2) **TMDb**: Enrich with popularity, votes, ratings via API; (3) **Wikipedia**: Extract plot summaries from sitelinks; (4) **Embeddings**: Generate dense/sparse embeddings using BGE-M3 with CLS token aggregation, parallel GPU processing.
 
-### Additional Files
+## Data Fields
 
-- **`knn_faiss_novelty.csv`**: K-nearest neighbors analysis results using FAISS
-  - Contains novelty scores and neighbor information for movies
+**Core**: `movie_id` (Wikidata QID), `title`, `year`, `imdb_id`, `tmdb_id`
 
-- **`umap_cluster_trajectories.png`**: Visualization of movie embeddings in UMAP space
-  - Shows temporal trajectories and genre clusters
+**Metadata**: `release_date`, `country`, `duration`, `wikidata_class`, `wikipedia_link`
 
-- **`concept_space/`**: Directory containing concept vocabulary and embeddings
-  - `concept_words_*.npy`: WordNet-based concept vocabulary (nouns)
-  - `concept_vecs_*.npy`: Pre-computed embeddings for concept words
-  - Used for concept extraction and semantic mapping
+**Creative Team**: `directors`, `directors_id`, `actors`, `actors_id` (pipe-separated for multiple)
 
-## Data Collection Pipeline
+**Genre**: `genre` (raw, comma-separated), `genre_id`, `genre_cluster_ids`, `genre_cluster_names` (processed)
 
-The dataset was created through a four-step pipeline:
-
-### Step 1: Wikidata Collection
-- Queries Wikidata SPARQL endpoint to fetch movie metadata
-- Filters for feature-length films (excludes TV series, short films, etc.)
-- Collects: titles, release years, genres, directors, actors, countries, Wikidata IDs
-- Year range: 1930-2024
-- Target: ~8,000 movies per year
-
-### Step 2: TMDb Enrichment
-- Enriches Wikidata entries with TMDb metadata via API lookup
-- Retrieves: popularity scores, vote counts, average ratings, TMDb IDs
-- Links movies using Wikidata IDs
-
-### Step 3: Wikipedia Plot Retrieval
-- Fetches full plot summaries from Wikipedia using Wikidata sitelinks
-- Handles redirects and extracts main plot sections
-- Filters out plots that are too short or contain insufficient narrative content
-- Stores plot text and section title
-
-### Step 4: Embedding Generation
-- Generates dense embeddings using BAAI/bge-m3 model
-- Uses CLS token aggregation method (default)
-- Supports multiple chunking strategies (CLS token, mean pooling, chunk-first, late chunking)
-- Generates sparse lexical weights for concept extraction
-- Processes embeddings in batches with parallel GPU support
+**Plot**: `plot` (full text), `plot_section`, `plot_length_chars`, `plot_length_tokens`, `num_different_tokens`, `token_shannon_entropy`
 
-## Data Fields
+**TMDb Metrics**: `popularity`, `vote_average`, `vote_count`
+
+**Financial** (if available): `budget`, `budget_currency`, `box_office`, `box_office_currency`, `box_office_worldwide`
 
-The `final_dataset.csv` file contains the following columns:
-
-### Core Identifiers
-- **`movie_id`**: Wikidata QID (e.g., "Q1931001") - primary identifier
-- **`title`**: Movie title
-- **`year`**: Release year (1930-2024)
-- **`imdb_id`**: IMDb identifier
-- **`tmdb_id`**: TMDb identifier
-
-### Metadata
-- **`release_date`**: Full release date
-- **`country`**: Production country
-- **`duration`**: Runtime in minutes
-- **`duration_all`**: All duration information
-- **`wikidata_class`**: Wikidata entity class
-- **`wikipedia_link`**: Link to Wikipedia article
-
-### Creative Team
-- **`directors`**: Director names (pipe-separated if multiple)
-- **`directors_id`**: Director Wikidata IDs (pipe-separated)
-- **`actors`**: Actor names (pipe-separated if multiple)
-- **`actors_id`**: Actor Wikidata IDs (pipe-separated)
-
-### Genre Information
-- **`genre`**: Raw genre strings (comma-separated)
-- **`genre_id`**: Genre Wikidata IDs (comma-separated)
-- **`genre_cluster_ids`**: Clustered genre IDs (after processing)
-- **`genre_cluster_names`**: Clustered genre names (after processing)
-
-### Plot Data
-- **`plot`**: Full plot summary text from Wikipedia
-- **`plot_section`**: Section title from which plot was extracted
-- **`plot_length_chars`**: Character count of plot
-- **`plot_length_tokens`**: Token count of plot
-- **`num_different_tokens`**: Number of unique tokens
-- **`token_shannon_entropy`**: Shannon entropy of token distribution
-
-### TMDb Metrics
-- **`popularity`**: TMDb popularity score
-- **`vote_average`**: Average user rating
-- **`vote_count`**: Number of votes
-
-### Financial Data (if available)
-- **`budget`**: Production budget
-- **`budget_currency`**: Budget currency
-- **`box_office`**: Box office revenue
-- **`box_office_currency`**: Box office currency
-- **`box_office_worldwide`**: Worldwide box office
-- **`box_office_worldwide_currency`**: Worldwide currency
-
-### Additional Metadata
-- **`set_in_period`**: Time period in which movie is set
-- **`awards`**: Awards information
-- **`summary`**: Additional summary (usually empty)
+**Other**: `set_in_period`, `awards`, `summary`
 
 ## Loading the Dataset
 
-### Python Example
+### Basic Python Example
 
 ```python
 import numpy as np
 import pandas as pd
 
-# Load movie metadata
+# Load metadata and embeddings
 df = pd.read_csv('final_dataset.csv', low_memory=False)
-print(f"Loaded {len(df)} movies")
-print(f"Year range: {df['year'].min()} to {df['year'].max()}")
-
-# Load dense embeddings
 embeddings = np.load('final_dense_embeddings.npy')
 movie_ids = np.load('final_dense_movie_ids.npy', allow_pickle=True)
 
-print(f"Embeddings shape: {embeddings.shape}")
-print(f"Number of movie IDs: {len(movie_ids)}")
-
-# Verify alignment: embeddings[i] corresponds to movie_ids[i]
-assert len(embeddings) == len(movie_ids)
-
 # Merge embeddings with metadata
-embeddings_df = pd.DataFrame({
-    'movie_id': movie_ids,
-    'embedding': list(embeddings)
-})
+embeddings_df = pd.DataFrame({'movie_id': movie_ids, 'embedding': list(embeddings)})
 combined_df = pd.merge(df, embeddings_df, on='movie_id', how='inner')
-print(f"Combined dataset: {len(combined_df)} movies with embeddings")
 ```
 
-### Using the Provided Utilities
-
-The dataset is designed to work with the utilities in `src/data_utils.py`:
+### Using Utilities
 
 ```python
-from src.data_utils import (
-    load_final_dataset,
-    load_final_dense_embeddings,
-    load_final_sparse_embeddings,
-    load_final_data_with_embeddings
-)
+from src.data_utils import load_final_data_with_embeddings
 
-# Load dataset with embeddings merged
-data_dir = "data/data_final"
 df = load_final_data_with_embeddings(
-    csv_path=f"{data_dir}/final_dataset.csv",
-    data_dir=data_dir,
-    verbose=True
-)
-
-# Access embeddings
-embeddings = np.array(df['embedding'].tolist())
-movie_ids = df['movie_id'].values
-```
-
-### Loading Sparse Embeddings (Lexical Weights)
-
-```python
-from src.data_utils import load_final_sparse_embeddings
-
-# Load sparse embeddings
-token_indices_list, weights_list, movie_ids = load_final_sparse_embeddings(
+    csv_path="data/data_final/final_dataset.csv",
     data_dir="data/data_final",
     verbose=True
 )
-
-# Access lexical weights for a specific movie
-movie_idx = 0
-token_indices = token_indices_list[movie_idx]  # Token IDs with non-zero weights
-weights = weights_list[movie_idx]  # Corresponding lexical weights
-movie_id = movie_ids[movie_idx]  # Wikidata QID
 ```
 
 ## Embedding Details
 
-### Model
-- **Model**: BAAI/bge-m3
-- **Embedding Dimension**: 1024
-- **Aggregation Method**: CLS token (default)
-- **Normalization**: L2-normalized embeddings
-
-### Alternative Chunking Methods
-The pipeline supports multiple chunking strategies (stored with suffixes):
-- `_cls_token`: CLS token aggregation (default)
-- `_mean_pooling`: Global mean pooling
-- `_chunk_first_{chunk_size}_{stride}`: Early chunking with specified parameters
-- `_late_chunking_{window_size}_{stride}`: Late chunking with specified parameters
-
-### Embedding Alignment
-- Embeddings are indexed by `movie_id` (Wikidata QID)
-- `final_dense_embeddings[i]` corresponds to `final_dense_movie_ids[i]`
-- To find embedding for a specific movie:
-  ```python
-  movie_id = "Q1931001"
-  idx = np.where(movie_ids == movie_id)[0][0]
-  embedding = embeddings[idx]
-  ```
+- **Model**: BAAI/bge-m3 | **Dimension**: 1024 | **Method**: CLS token aggregation | **Normalization**: L2-normalized
+- **Alignment**: `final_dense_embeddings[i]` corresponds to `final_dense_movie_ids[i]`
+- **Alternative chunking**: Supports mean pooling, chunk-first, late chunking (stored with suffixes)
 
 ## Concept Space
 
-The `concept_space/` directory contains pre-computed concept vocabularies and embeddings:
-
-- **Source**: WordNet noun synsets
-- **Filtering**: Zipf frequency ≥ 4.0 (common words)
-- **Size**: Top 20,000 most frequent nouns
-- **Embedding Model**: BGE-small-en-v1.5 (L2-normalized)
-- **Usage**: Concept extraction from movie plots using lexical weights
-
-Files are parameterized by:
-- Zipf threshold
-- Vocabulary size
-- Model name
-
-Example: `concept_words_zipf2.5_vocab10000_BAAI_bge-m3.npy`
+Pre-computed WordNet-based concept vocabulary: top 20K nouns (Zipf ≥ 4.0), embedded with BGE-small-en-v1.5. Files parameterized by Zipf threshold, vocabulary size, and model name.
 
 ## Data Quality
 
-### Completeness
-- All movies have: `movie_id`, `title`, `year`, `wikipedia_link`
-- ~92,000 movies have plot summaries
-- ~91,000 movies have TMDb metadata (popularity, votes)
-- ~75,000 movies have director information
-- ~81,000 movies have genre information
+**Completeness**: All movies have core fields; ~92K plots, ~91K TMDb data, ~75K directors, ~81K genres.
 
-### Filtering
-- Only feature-length films (excludes TV series, short films, video games)
-- Plots filtered for minimum length and narrative content
-- Non-movie entries removed based on Wikidata class
+**Filtering**: Feature films only; plots filtered by length/entropy; explicit content excluded; duplicates removed.
 
-### Data Cleaning
-- Plot text cleaned (removed line breaks, normalized whitespace)
-- Genre strings normalized and clustered
-- Duplicate entries removed based on movie_id
+**Cleaning**: Plot text normalized; genres clustered; Shannon entropy threshold: 4.8398.
 
 ## Use Cases
 
-### 1. Temporal Semantic Drift Analysis
-Analyze how movie plot semantics evolve over time:
-```python
-# Group by decade and compute centroid shifts
-decades = (df['year'] // 10) * 10
-decade_embeddings = df.groupby(decades)['embedding'].apply(
-    lambda x: np.mean(np.array(x.tolist()), axis=0)
-)
-```
+**Temporal Analysis**: Compute decade centroids to analyze semantic drift over time.
 
-### 2. Genre Classification
-Use embeddings for genre prediction or clustering:
-```python
-from sklearn.cluster import KMeans
-
-embeddings_array = np.array(df['embedding'].tolist())
-kmeans = KMeans(n_clusters=10)
-clusters = kmeans.fit_predict(embeddings_array)
-```
-
-### 3. Movie Similarity Search
-Find similar movies using cosine similarity:
-```python
-from sklearn.metrics.pairwise import cosine_similarity
+**Genre Classification**: Use embeddings for clustering or classification tasks.
 
-target_embedding = embeddings[0].reshape(1, -1)
-similarities = cosine_similarity(target_embedding, embeddings)
-similar_movies = np.argsort(similarities[0])[-10:][::-1]
-```
+**Similarity Search**: Find similar movies using cosine similarity on embeddings.
 
-### 4. Concept Extraction
-Extract semantic concepts from plots using lexical weights:
-```python
-# Use sparse embeddings to map plot nouns to concept space
-# See src/analysis/concept_extraction.py for implementation
-```
+**Concept Extraction**: Map plot nouns to concept space using sparse lexical weights.
 
 ## Citation
 
-If you use this dataset, please cite:
-
 ```bibtex
 @dataset{movie_plot_embeddings_2026,
   title={Plot Twists Over Time: How Movie Stories Have Changed Over 95 Years},
@@ -373,57 +159,25 @@ If you use this dataset, please cite:
 
 ## License
 
-This dataset is derived from multiple open data sources with different licenses. Users must comply with all applicable source licenses:
-
-- **Wikidata**: Released under [CC0 1.0 Universal (Public Domain)](https://creativecommons.org/publicdomain/zero/1.0/). No restrictions on use.
-- **Wikipedia**: Content is licensed under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/). Requires attribution and share-alike terms.
-- **TMDb**: Data is licensed under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) for non-commercial use. Commercial use requires separate licensing from TMDb.
+This dataset inherits licenses from source data:
+- **Wikidata**: CC0 1.0 (Public Domain)
+- **Wikipedia**: CC BY-SA 4.0
+- **TMDb**: CC BY-NC 4.0 (non-commercial)
 
-**Dataset License**: This aggregated dataset inherits the most restrictive license from its sources. For non-commercial research purposes, this dataset may be used under CC BY-NC 4.0 with proper attribution to all sources. For commercial use, users must obtain appropriate licenses from TMDb and comply with Wikipedia's CC BY-SA 4.0 terms.
+**Dataset License**: CC BY-NC 4.0 for non-commercial research. Commercial use requires TMDb licensing and Wikipedia compliance.
 
-**Code/Software**: The code used to create this dataset is licensed under MIT License (see main repository LICENSE file).
+**Code**: MIT License (see main repository).
 
-**Attribution**: When using this dataset, please cite:
-- Wikidata contributors
-- Wikipedia contributors  
-- TMDb (The Movie Database)
-- This dataset (see Citation section)
+**Attribution Required**: Cite Wikidata, Wikipedia, TMDb contributors, and this dataset.
 
 ## Acknowledgments
 
-- **Wikidata**: Movie metadata and identifiers
-- **TMDb**: Popularity scores, ratings, and additional metadata
-- **Wikipedia**: Plot summaries and descriptions
-- **BAAI**: BGE-M3 embedding model
-- **WordNet**: Concept vocabulary for semantic analysis
-
-## Contact
-
-For questions or issues, please refer to the main repository: [GroupDataLiteracy](https://github.com/your-repo/GroupDataLiteracy)
-
-## Dataset Statistics
-
-- **Total Movies**: ~92,000
-- **Year Range**: 1930-2024
-- **Movies with Plots**: ~92,000
-- **Movies with TMDb Data**: ~91,000
-- **Movies with Genres**: ~81,000
-- **Embedding Dimension**: 1024
-- **Average Plot Length**: ~1,500 characters
-- **Unique Genres**: ~50 (after clustering)
-
-## File Sizes (Approximate)
-
-- `final_dataset.csv`: ~200 MB
-- `final_dense_embeddings.npy`: ~380 MB (for 92K movies × 1024 dims)
-- `final_dense_movie_ids.npy`: ~7 MB
-- `final_sparse_embeddings.npz`: ~500 MB (variable, depends on sparsity)
+Wikidata, TMDb, Wikipedia, BAAI (BGE-M3), WordNet
 
 ## Notes
 
-- All embeddings are L2-normalized
-- Movie IDs are Wikidata QIDs (strings starting with "Q")
-- Plot text is cleaned and normalized
-- Genre information may contain multiple genres per movie (pipe-separated)
-- Some fields may be NaN for older movies or less popular films
-- The dataset is designed for research in temporal semantic analysis and embedding-based movie analysis
+- Embeddings are L2-normalized
+- Movie IDs are Wikidata QIDs (format: "Q####")
+- Plot text cleaned and normalized
+- Genres may be multi-label (pipe-separated)
+- Some fields may be NaN for older/less popular films