如何根据某人之前看过的电影推荐电影？

Question

For a machine learning exercise I am working on, I am given a dataset where each row contains the following features:

• the person's name,
• age,
• gender, and
• the movie they watched.

My task is to suggest other movies that the person might like based off these features.

The thing is, I am not given a feature set for movies. I am only given the dataset described above.

I already know I need to generate a feature set for movies. However, I don't know how to approach this.

After I create the feature set, I will convert each movie's feature set into an embedding (vector). Then I will use a similarity-matching library (such as Spotify's Annoy ) to find return embeddings of similar movies.

The part I am stuck at is how I can use the dataset to generate a feature set for each movie.

Answer 1

Imagine that you have a table like this:

+-------+-----+--------+---------------------+
| Name  | Age | Gender |        Movie        |
+-------+-----+--------+---------------------+
| John  |  23 | Male   | John the Ripper     |
| Luke  |  18 | Male   | The Star Wars       |
| Ann   |  18 | Female | Mr. Nobody          |
| Alice |  12 | Female | Alice in Wonderland |
| Bruce |  64 | Male   | Armageddon          |
+-------+-----+--------+---------------------+

I. First of all, you need to separate this table by two parts:

1. The features vector which contains Name , Age , Gender columns.
2. The purpose vector which contains only Movie column.

II. After that you could to encode your strings into numbers:

1. The column Name will be encoded into unique index.
2. The column Age will not changes.
3. The column Gender will be encoded into binary values (0, 1).
4. The column Movie will be encoded into unique index values.

For example:

+------+-----+--------+-------+
| Name | Age | Gender | Movie |
+------+-----+--------+-------+
|    0 |  23 |      1 |     3 |
|    1 |  18 |      1 |     2 |
|    2 |  18 |      0 |     4 |
|    3 |  12 |      0 |     1 |
|    4 |  64 |      1 |     0 |
+------+-----+--------+-------+

III. Then you may separate your vector on two parts:

1. Train data for machine learning algorithm feeding ( rows 1:3 ).
2. Test data for contest ML algorithm that you feed ( rows 3:5 ).

The proportion between this separate set may be different, but usually train data set picks greater than test data set.

IV. Sometimes you could need for scaling your data.

For example:

+------+--------+--------+-------+
| Name |  Age   | Gender | Movie |
+------+--------+--------+-------+
| 0.0  | 0.3594 |      1 | 0.6   |
| 0.2  | 0.2813 |      1 | 0.4   |
| 0.4  | 0.2813 |      0 | 0.8   |
| 0.6  | 0.1875 |      0 | 0.2   |
| 0.8  | 1.0000 |      1 | 0.0   |
+------+--------+--------+-------+

In this sample after steps I-IV you will get:

feature_train = [[ 0.0, 0.3594, 1 ], [ 0.2, 0.2813, 1 ], [ 0.4, 0.2813, 0 ]]
purpose_train = [ 0.6, 0.4, 0.8 ]
feature_test  = [[ 0.6, 0.1875, 0], [0.8, 1.0000, 1]]
purpose_test  = [[ 0.2, 0.0]]

That's all to prepare data in simple way.

[UDP]

After all this steps, you should teach your algorithm by the data, and then you may predict the favorite Movie by Name, Age and Gender for the choosed one.