This document provides more detail about the sample movie recommender. You can learn more about building recommenders with GraphLab Create in our user guide.
The movie_recommender.py script performs the following actions:
Load Data -> Prepare Data -> Train Model -> Explore Model
For this sample, two datasets are loaded: movie item data (27,279 movies), and user action data (1,000,000 ratings).
Movie item data describes individual movies, with fields such as title, description, tags, etc.:
| movieId | title | genres |
|---|---|---|
| 1 | Toy Story (1995) | Adventure,Animation,Children,Comedy,Fantasy |
| 2 | Jumanji (1995) | Adventure,Children,Fantasy |
| 3 | Grumpier Old Men (1995) | Comedy,Romance |
User action data documents past interactions between users and movie items, and
includes fields such as user_id, item_id, and a numeric rating score column (optional):
| userId | movieId | rating | timestamp |
|---|---|---|---|
| 1 | 2 | 3.5 | 1112486027 |
| 1 | 29 | 3.5 | 1112484676 |
| 1 | 32 | 3.5 | 1112484819 |
Once the data is loaded, you typically want to clean it up or transform it in various ways. Here, we will remove items that were rated less than 5 times, because sometimes infrequently rated items can cause the model to behave unpredictably.
We also extract the year and title from the original title column, which
combined these together, and parse the genres column by splitting on commas.
See our user guide on feature engineering for more discussion.
Next, we randomly split the data into a training set and validation set.
We use gl.recommender.create
to build a recommender model on the training set with default settings.
Once you've trained the model, you can use it to make recommendations or to find similar items. The sample script launches an interactive web-based view for exploring and evaluating the model:
Find more information about how to use your model in the user guide.
You can replace the sample data provided in this project with your own
dataset. The easiest way to do this is to look at the items.csv and
ratings.csv file loaded by the movie_recommender.py script and format your data
the same way. Then, simply replace the filenames in the gl.SFrame.read_csv
commands in movie_recommender.py with your own data filenames.
If you want to use different column names in your data files, you will
also need to make other changes to the example script. For example,
if your data contains a bookId column instead of a movieId column,
you will need to replace references to movieId in movie_recommender.py.
Once you have built an initial model from your own data, you might also
find that the model performance can be improved if you change or replace the
data cleaning and feature extraction commands in movie_recommender.py with customized instructions.
The ideal set of data transformations typically varies from dataset to dataset,
so you will have to experiment a bit. For some suggestions,
see our user guide on feature engineering.