Recomputing Preference Engine predictions

The <eng_instance_name>.engine.titan.recomputation_bound configuration parameter specifies the percentage change allowed in a user's ratings before the LikeMindsserver recomputes the user's predictions.

For example:

music_pref.engine.titan.recomputation_bound = 10.0

Ordinarily, the LikeMindsserver generates predictions based on a user's mentors, which the sifter computes and makes available to the database. When a user has no mentors, perhaps because he or she has just arrived at the site, or when the user's ratings or transactions have changed beyond the percentage specified here, the LikeMindsserver selects mentors from a reduced set of candidates and recomputes the user's predictions.

Use this setting with caution, as selecting mentors is a relatively expensive operation: a low percentage setting can lead to excessive CPU load with little or no gain in prediction quality. A higher setting will improve performance, but predictions may be less accurate.

Parent: User predictability