ansat7/hangman_english

Hangman Movie Titles Dataset

Dataset Summary

This dataset contains cleaned English movie titles for Hangman-style letter guessing. The task is to predict missing A-Z letters from a partially revealed movie title.

The final title list combines TMDB-based sources, then applies normalization, filtering, deduplication, and a fixed train/validation/test split.

Dataset Structure

Files

File	Rows	Description
hangman_dataset.csv	619,708	Full cleaned title list
hangman_train.csv	495,131	Training split
hangman_val.csv	61,891	Validation split
hangman_test.csv	61,892	Test split
index.csv	178,181	Single-word lookup list for late-game fallback

Data Fields

Title split files:

Column	Type	Description
title	string	Uppercase movie title using A-Z and spaces

Lookup file:

Column	Type	Description
word	string	Uppercase single word used for candidate lookup

Data Splits

The title dataset is split with a fixed seed into 80/10/10 train, validation, and test splits.

Split	Rows
Train	495,131
Validation	61,891
Test	61,892

Preprocessing

• Kept English movie titles.
• Removed empty titles and titles with numbers.
• Normalized text to uppercase ASCII.
• Kept only A-Z and spaces.
• Collapsed repeated spaces.
• Deduplicated titles.
• Built index.csv as a single-word lookup list from title words and supporting word sources.

Intended Use

This dataset is intended for training and evaluating Hangman solvers over English movie titles. A game state can be generated by hiding some letters, revealing at least one letter per word, and adding already-missed guesses.

Limitations

• Numbers, punctuation, symbols, and stylized spellings are removed.
• The dataset focuses on English movie titles, but some titles still contain names, acronyms, transliterated words, or rare title-specific tokens.
• index.csv is a helper lookup file, not the main supervised training target.

Sources

• TMDB movie-title data
• TMDB daily export data