Microsoft is seeking to address one of the most frustrating problems for Outlook users, which is the difficulty of finding specific emails amid a flood of irrelevant results. This update redesigns the search mechanism by improving result ranking, raising performance efficiency, and integrating artificial intelligence technologies more intelligently to reduce time wasted on manual searching.
Improvements in Ranking Mechanism and Word Processing
The new updates focus on processing "stop words" such as "the", "and", and "is", which previously caused inaccurate results that relied solely on chronological ordering. Seven ranked results will also be displayed instead of three, as the new ranking process relies on multiple factors including message recency, sender identity, and the extent of interaction with the message, ensuring that the user reaches the required information more quickly.
Reliance on Small Language Model (SLM) Technology
Technically, the new ranking system relies on a Small Language Model developed specifically for the Outlook environment. This model was built through a process known as "distillation", where it is trained under the supervision of a larger language model. This strategy aims to make the search engine lightweight and highly efficient in classifying results without consuming massive computational resources, balancing high accuracy with response speed.
What is Practically Changing for Users?
- Improved Performance in Weak Connectivity: In the Outlook Classic version, the application will automatically rely on locally indexed content when internet connectivity is weak, a feature that will later arrive for the new Outlook version.
- Reduced Latency: The improvements have successfully reduced latency, especially in searches that were classified among the slowest by 5%.
- Direct Feedback Loop: Users will be able to rate search results, providing Microsoft with real-world data to improve ranking algorithms over time.