Seeing different detection results?
As our customers begin to upgrade from Version 3 of the 51Degrees service to Version 4, there are certain differences that may be seen between the two. This blog will go through some of the differences you can expect to see.
Segmenting the User-Agent into substrings
One of the biggest differences between Version 3 and Version 4 is within the component (Hardware, Software, Browser, and Crawler) identification.
The User-Agent string contains information on the device, which can be divided out into separate segments. For example, the following Google Pixel 6 User-Agent can be divided into six segments (or “substrings”):
Mozilla/5.0 (Linux; Android 12; Pixel 6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/93.0.4577.62 Mobile Safari/537.36
Mozilla/5.0
(Linux; Android 12;
Pixel 6)
AppleWebKit/537.36
(KHTML, like Gecko)
Chrome/93.0.4577.62 Mobile Safari/537.36
Within V3, the whole User-Agent string is used to identify the device in one single detection. However, V4 looks at the separate segments of the User-Agent string and performs multiple detections to identify the corresponding components.
Sometimes, we see unusual and rare User-Agents. In this case, V3 would struggle to detect a User-Agent of this nature, as it hasn’t been seen before and is therefore unable to match the User-Agent to a device.
V4 mitigates this problem, as it can look at the separate substrings of the User Agent string and deduce the device components from there. Irrelevant information within the string isn’t considered – we discuss this further in our documentation.
Because of this difference in substring detection, you may see conflicting results between V3 and V4 for the same User-Agent. However, you can have more confidence in the accuracy of Version 4 when detecting new or unusual User-Agent strings.
Machine learning algorithms
Version 3 and Version 4 use the same data but have different machine learning algorithms.
We continually update the V4 algorithm with the latest changes from the industry. When Google announced User-Agent Client Hints, their replacement for the User-Agent string, we were able to update V4 to contain full User-Agent Client Hints support.
Algorithm changes are a normal and expected part of any service improvement.
Enabling usage sharing
As part of the Version 4 machine learning algorithm, we encourage our users to enable usage sharing.
When data is sent from our users to our machine learning algorithm, we can collect real User-Agents and UA-CH information that circulate the web. This information is then fed into our algorithm, ensuring the detection is as accurate and up to date as possible.
If you are experiencing a drop in detection accuracy with some unusual User Agents, we encourage you to check usage sharing is enabled within your configuration.
Upgrading from Version 3 to Version 4
As the V4 machine learning algorithm grows, you can expect performance differences between Version 3 and Version 4. It’s worth noting that these differences are not necessarily a cause for concern.
This does not mean that V3 performs poorly, but rather the accuracy and relevancy of V4 improves over time, increasing the accuracy gap between V3 and V4. V3 accuracy has plateaued, whereas V4 can only get better.
All the above improvements from Version 3 to Version 4 result in V4 being faster and more accurate for device detection.
Like the sound of upgrading to Version 4? Get in touch with our Sales team to find out more.
In the meantime, you can read our migration guide for your specific programming language or look at our pricing page and work out which solution works best for you.