How's user behavior, input, and feedback incorporated into the model?

Machine learning models incorporate user behavior and input by ingesting new or changed data to train the model every day. By continuously retraining, the model learns from use and delivers best results.

Note: Your organization's feedback data is isolated, and not incorporated into shared models.

Oracle also gathers telemetry data about product usage to incorporate user behavior and input, for example:

• We evaluate telemetry metrics to determine the data features and model configurations needed to optimize outputs. Telemetry data is also used to identify patterns that might lead to model or feature enhancements.
• Telemetry data is also used to identify changes in user behavior. We adapt model weightings to optimize outputs, based on the real-world data patterns that evolving user behavior displays.
• Telemetry data is also used to promote or relegate some outputs. For example, if a recommendation is liked, accepted, or used, it may be given a positive score, and if not, a negative one. We also monitor indicators that are more passive than clicking a Like button. For example, the number of changes to an autogenerated response, or the time spent on the task used to derive an implicit like or dislike.

The large language model (LLM) that Oracle Fusion Applications Cloud uses is a foundational model that's not further trained or fine tuned. But, we do monitor the LLM's performance by assessing the LLM's output and storing a score. We also record whether the user accepts or changes the LLM's output, which indicates the quality of the LLM's response from the user's perspective. We evaluate the model's performance when considering model upgrades.