Kaggle Movies

About the dataset

This dataset is a subset of the Kaggle Movies Dataset, containing over 40,000 movie titles and overviews. It also includes pre-computed 512-dimensional vector embeddings (generated with OpenAI's text-embedding-3-small model) for both the title and overview fields, making it useful for experimenting with semantic search in MotherDuck.

How to query the dataset

This dataset is available as part of the sample_data database, which is automatically attached to every MotherDuck account.

Schema

Column Name	Column Type	Description
title	VARCHAR	Movie title
overview	VARCHAR	Short description or synopsis of the movie
title_embeddings	FLOAT[512]	Pre-computed vector embedding of the title
overview_embeddings	FLOAT[512]	Pre-computed vector embedding of the overview

Example queries

Browse movies

SELECT title, overview
FROM sample_data.kaggle.movies
LIMIT 10;

Find similar movies using vector search

Use the pre-computed embeddings together with the embedding function to find movies similar to a search query:

SELECT
    title,
    overview,
    array_cosine_similarity(
        overview_embeddings,
        embedding('a space adventure with aliens')
    ) AS similarity
FROM sample_data.kaggle.movies
WHERE overview IS NOT NULL
ORDER BY similarity DESC
LIMIT 10;

Find movies similar to another movie

WITH target AS (
    SELECT overview_embeddings
    FROM sample_data.kaggle.movies
    WHERE title = 'The Matrix'
    LIMIT 1
)
SELECT
    m.title,
    m.overview,
    array_cosine_similarity(m.overview_embeddings, t.overview_embeddings) AS similarity
FROM sample_data.kaggle.movies m, target t
WHERE m.title != 'The Matrix'
ORDER BY similarity DESC
LIMIT 10;