3.11.8 Generating Recommendations for Similar Users

This section describes how to generate recommendations for similar users.

• For known users, the output includes a list of predicted users that have similar behavior and taste.

• The predictions are expressed in cosine similarity, and range from 0, very dissimilar, to 1, very similar.

• For a new user, there is no information to provide a prediction. This will produce an error.

Before You Begin

1. Connect to your HeatWave DB System.

2. Complete the steps for training the recommendation model. See Section 3.11.2, “Training a Recommendation Model”.

3. Review Section 3.11.3, “Using a Recommendation Model”.

Generating Similar Users

To generate recommendations for similar users:

1. Use the ML_MODEL_LOAD routine to load the trained model with recommendations. For example:

   mysql> CALL sys.ML_MODEL_LOAD(@model, NULL);

2. Run the ML_PREDICT_TABLE or ML_PREDICT_ROW routine to generate predictions. You must set the recommend option to users_to_users.

   The following example uses explicit feedback and runs the ML_PREDICT_ROW routine to predict the top three users similar to another user.

   mysql> SELECT sys.ML_PREDICT_ROW('{"user_id": "846"}', @model,  
             JSON_OBJECT("recommend", "users_to_users", "topk", 3));
   +--------------------------------------------------------------------------------------------------------------------------------+
   | sys.ML_PREDICT_ROW('{"user_id": "846"}', @model,  JSON_OBJECT("recommend", "users_to_users", "topk", 3))                       |
   +--------------------------------------------------------------------------------------------------------------------------------+
   | {"user_id": "846", "ml_results": "{"predictions": {"user_id": ["62", "643", "172"], "similarity": [0.7413, 0.7275, 0.6709]}}"} |
   +--------------------------------------------------------------------------------------------------------------------------------+
   1 row in set (0.2373 sec)

   • User 846 is predicted to be most similar to users 62, 643, and 172.

   • For user 846, user 62 has similarity value of 0.7413, user 643 has a similarity value of 0.7275, and user 172 has a similarity value of 0.6709.

   The following example uses implicit feedback and runs the ML_PREDICT_TABLE routine to predict the top three users similar to other users.

   mysql> CALL sys.ML_PREDICT_TABLE('mlcorpus.test_sample', @model, 'mlcorpus.users_to_users_recommendation',  JSON_OBJECT("recommend", "users_to_users", "topk", 3));
   Query OK, 0 rows affected (3.4959 sec)
   
   mysql> SELECT * FROM mlcorpus.users_to_users_recommendation LIMIT 3;
   +-------------------+---------+---------+--------+------------------------------------------------------------------------------------------------+
   | _4aad19ca6e_pk_id | user_id | item_id | rating | ml_results                                                                                     |
   +-------------------+---------+---------+--------+------------------------------------------------------------------------------------------------+
   |                 1 | 1026    | 13763   |      1 | {"predictions": {"user_id": ["2215", "1992", "4590"], "similarity": [1.0, 1.0, 1.0]}}          |
   |                 2 | 992     | 16114   |      1 | {"predictions": {"user_id": ["2092", "1809", "5174"], "similarity": [0.8576, 0.8068, 0.7533]}} |
   |                 3 | 1863    | 4527    |      1 | {"predictions": {"user_id": ["4895", "1405", "2972"], "similarity": [0.9845, 0.8666, 0.8623]}} |
   +-------------------+---------+---------+--------+------------------------------------------------------------------------------------------------+
   3 rows in set (0.0004 sec)

   • Users 2215, 1992, and 4590 are predicted to be most similar to user 1026.

   • Users 2092, 1809, and 5174 are predicted to be most similar to user 992.

   • Users 4895, 1405, and 2972 are predicted to be most similar to user 1863.

   • The similarity values for each item-item pair are in ml_results after similarity. For example, for user 1026, users 2215, 1992, and 4590 have a similarity value of 1.0.

   The values in columns item_id and rating can be disregarded as they are part of the data that was trained from the input table.