To improve the efficiency of personal assistant AI systems, researchers at Apple have developed a new AI framework that leverages user engagement signals to automatically label data.

Personal assistant AI systems are widely used these days as a means to accomplish tasks through natural language commands. However, these systems generally rely on supervised machine learning algorithms. These algorithms require large amounts of hand-annotated training data which is expensive and time-consuming to collect. 

For programmers and researchers dealing with machine learning, it will be highly beneficial if they can incorporate unsupervised, weakly supervised, or distantly supervised data. The new AI framework from Apple tries to solve this bottleneck by leveraging user engagement signals (user behaviours that demonstrate a positive or negative response to content) to create granular entity labels for training data augmentation automatically. 

In a pre-print research paper titled " Leveraging User Engagement Signals For Entity Labeling in a Virtual Assistant," researchers notes that their "results show that learning from data automatically labelled by user engagement signals achieves significant accuracy gains in a production deep learning system, when measured on both the sequence labelling task as well as on user-facing results produced by the system end-to-end."

The authors of the paper write that they "believe this is the first use of user engagement signals to help generate training data for a sequence labelling task on a large scale and can be applied in practical settings to speed up new feature deployment when little human-annotated data is available."