Ishaank18/screenplay-features

Screenplay Scene Salience Features

Pre-extracted linguistic and narrative features for screenplay scene salience detection from the MENSA dataset.

Dataset Description

This dataset contains 913 linguistic features extracted from movie screenplays in the MENSA dataset. Features are organized into 24 feature groups covering various aspects of linguistic, narrative, and discourse analysis.

Dataset Statistics

Split	Samples	Size
Train	117,503	172.9 MB
Validation	8,052	16.1 MB
Test	8,156	16.1 MB
Total	133,711	140.1 MB

Feature Groups (24 groups)

• base
• bert_surprisal
• character_arcs
• emotional
• gc_academic
• gc_basic
• gc_char_diversity
• gc_concreteness
• gc_dialogue
• gc_discourse
• gc_narrative
• gc_polarity
• gc_pos
• gc_pronouns
• gc_punctuation
• gc_readability
• gc_syntax
• gc_temporal
• ngram
• ngram_surprisal
• plot_shifts
• rst
• structure
• surprisal

Usage

Option 1: Load with Hugging Face datasets (Recommended)

from datasets import load_dataset

# Load a single feature group
ds = load_dataset("Ishaank18/screenplay-features", data_files="train/base.parquet")
df = ds['train'].to_pandas()

# Load multiple groups for training
ds = load_dataset("Ishaank18/screenplay-features", 
                  data_files={
                      "train": ["train/base.parquet", "train/gc_polarity.parquet", "train/emotional.parquet"]
                  })
df = ds['train'].to_pandas()

# Load all splits for evaluation
ds = load_dataset("Ishaank18/screenplay-features",
                  data_files={
                      "train": "train/gc_polarity.parquet",
                      "validation": "validation/gc_polarity.parquet",
                      "test": "test/gc_polarity.parquet"
                  })

Option 2: Load with pandas directly

import pandas as pd

# From HuggingFace URL
df = pd.read_parquet("hf://datasets/Ishaank18/screenplay-features/train/base.parquet")

# Or if you have the repo cloned locally
df = pd.read_parquet("train/base.parquet")

Option 3: Use custom loader (Easiest)

from feature_cache.load_hf import load_groups

# Load features and labels
X, y = load_groups(
    groups=["base", "gc_polarity", "emotional", "rst"],
    split="train",
    hf_repo="Ishaank18/screenplay-features"
)

# Load features only (no labels)
X = load_groups(
    groups=["base", "gc_polarity"],
    split="test",
    include_label=False,
    hf_repo="Ishaank18/screenplay-features"
)

Data Structure

Each parquet file contains:

• movie_id (string): Unique movie identifier
• scene_index (int): Scene index within the movie (0-indexed)
• label (int): Salience label
  • 0 = Non-salient scene
  • 1 = Salient scene
• Feature columns: Various linguistic/narrative features (float/int)

Example row structure:

movie_id	scene_index	label	feature_1	feature_2	...
tt0111161	42	1	0.85	12.3	...

Feature Categories

The features are organized into the following categories:

Base Features

• Basic linguistic statistics (token count, sentence count, etc.)
• Structural position features (act, scene positions)

GenreClassifier (GC) Features

• gc_basic: Basic linguistic metrics
• gc_char_diversity: Character diversity metrics
• gc_concreteness: Concreteness scores
• gc_dialogue: Dialogue-specific features
• gc_discourse: Discourse markers and connectives
• gc_narrative: Narrative structure features
• gc_polarity: Sentiment polarity scores
• gc_pos: Part-of-speech distributions
• gc_pronouns: Pronoun usage patterns
• gc_punctuation: Punctuation statistics
• gc_readability: Readability metrics
• gc_syntax: Syntactic complexity features
• gc_temporal: Temporal expressions

Narrative Features

• character_arcs: Character development metrics
• plot_shifts: Plot progression indicators
• structure: Narrative structure features
• emotional: Emotional arc features

Linguistic Features

• ngram: N-gram diversity metrics
• rst: Rhetorical Structure Theory features
• bert_surprisal: BERT-based surprisal scores
• ngram_surprisal: N-gram-based surprisal