Detect mobile devices from user agent string [closed]

Question

I am looking for a way to analyze user agent strings to determine whether they were generated by mobile devices. This needs to be java-based and usable in large batch log file analysis on hadoop for generating statistics (i.e., a web service wouldn't be appropriate).

I have seen WURFL, but given that I just need a binary mobile/not mobile response, the license fee seems prohibitive.

So far I have been using UADetector, which is almost exactly what I need. However, I have encountered some limitations with it. In my testing, I have found many user agent strings that provide enough information to determine that the user agent is from a mobile device, but are reported as UNKNOWN by UADetector.

For example, poorly-standardized Android apps can send the UA string "Android". This is enough to know that it came from a mobile device, but UADetector reports this UserAgentType as UNKNOWN rather than MOBILE_BROWSER.

Apache Mobile Filter's Lite Device Detection does the right thing, but I need something I can use from Java.

Can anyone recommend a better solution?

Answer 1

I'm the founder and maintainer of the MobileESP project, a free open source cross-platform library for detecting mobile devices. It's still very much alive! :-)

www.mobileesp.org

MobileESP only gives binary "is mobile" responses. You can detect by platform like iOS, Android or Windows Phone, or by device category, like "iPhone Tier" smartphones vs. tablet. Be sure to take a quick review of the API page.

As you may know, useragent strings vary widely. If the browser shipped on the device, the manufacturer may customize it. For example, HTC often customizes the native Android browser's useragent string.

Google provides recommendations on how the OEM should customize the useragent. If the device should be considered a phone, then Google recommends including the word "mobile" element in the string. But if the device should be considered a tablet, then the string should not contain "mobile." Adherence to this recommendation varies widely, of course.

Third party browsers like Opera or Maxthon can put whatever they want to in the useragent string -- and do! Certain "new" browsers which shall remain nameless have been doing very poor jobs of putting the correct information in their useragent strings for each platform (e.g., Android vs. iOS versions). There's not much you can do unless you get a lot of traffic from these browsers and wish to invest in tracking their exact useragent values per platform and software rev.

Anyway, MobileESP was created with the vision of doing the detection on a page-by-page basis when the page is served. I purposefully wrote the code to be very easy to read and customize, too.

To do the batch processing, you might do something like this:

1.) In the constructor, comment out the initDeviceScan() method. You won't need this for bulk processing.

2.) Pass the UserAgent and an empty string in to the constructor (UAgentInfo()).

3.) Then run whatever detect methods you're interested in. Be thoughtful about the order in which you do them to save time, based on a scan of your users.

For example, if most of your users are on iPhone and that's one of the detection criteria you're interested in, then run that check first. If this example, you certainly wouldn't run the BlackBerry method first!

My contact info is in the source code and on the web site. Send me a note if you have any questions or run into any bugs. Definitely look around the MobileESP.org web site for some tips.

Best wishes on your project, Aniket!

• Anthony