Apple explains how automatic review summaries work in the App Store in iOS 18.4
With the release of iOS 18.4, Apple has added a new feature to the App Store that creates a summary of user reviews of apps and games.
Here's What We Know
In a new post on its machine learning blog, the company shared details about how the feature works.
Apple uses a multi-stage Large Language Model (LLM) system to create summaries, which is designed to provide balanced and accurate reviews that reflect the opinions of users. The company notes that it prioritises "safety, fairness, truthfulness, and usefulness" when creating summaries.
One of the main challenges is that reviews change due to new app updates, bug fixes, or new features. Therefore, the results must be dynamically adapted to remain relevant. In addition, the system must effectively handle both short and long reviews, filtering out unnecessary comments or spam.
To begin with, the LLM system ignores reviews that contain spam, foul language or fraud. Other reviews are processed through several stages, where the system identifies the main ideas, balances positive and negative opinions, and generates a summary of 100 to 300 characters in length.
During the development of the feature, thousands of summaries were reviewed by human experts to assess the usefulness, structure, and security of the results.
Apple's full blog post gives a more detailed insight into each step of the process and may be of interest to those who want to learn more about the company's approach to using large-scale language models.
Source: Apple
