So, Ive heard about a Facebook mobile app update that is freaking some people out (posted here: https://newsroom.fb/news/2014/05/a-new-optional-way-to-share-and-discover-music-tv-and-movies/) For those of you who are thinking TL;DR, heres the short version: Facebook is going to add the option to turn on a Shazam-like feature when you post status updates from your mobile app. What this means is itll make a very short (Shazam uses about 10 seconds) temporary recording of the audio it hears, and then itll turn that makes a fingerprint of that data, then compares it to the fingerprints in its database. These fingerprints arent actually the sounds it hears, but rather it takes the audio, converts it to a spectrogram, then looks at that for key points. It then takes one of those key ponts, and gets its relation to a set of other key points, and turns that data it into a a hash. Hashing is a kind of encoding that takes an amount of data and turns it into a much smaller set of numbers. Then is searches its database for that same hash. If it finds it, it tells you what song youre listening to. The only data that ever leaves your phone is that hash number, all that converting and checking is done on the phone. This is because 10 seconds of audio is a HUGE amount of data to send over a wireless network. A mono CD quality MP3 of that 10 second clip is about 882 KB, where as the average Shazam fingerprint is about 1.4 KB. As you can imagine, this tiny amount of data isnt super accurate. Shazams accuracy is around 60%, and its the best on the market and has been gathering data for 10 years. The way the petition describes the feature is that it will listen to your conversations and I guess give said data to the NSA, or McDonalds, or whomever. Not only is this wrong, its practically impossible. Facebook gets 55 milion status updates a day. The average time spent on Facebooks mobile app in a day is about 3x the amount spent on the full site, but as I personally prefer to do my status from a computer, Ill say that only 5/8ths of the posts are done via mobile, so thats only 34 million posts. If Facebook did want full audio of what you were saying everytime you posted, thatd be about 10 years worth of time per day, and almost 28 Terabytes of data. And audio data is hard as heck to search. Our best audio mining tools can only speed that up by a couple of hundred times, but even at 1000x, that 10 years will take about 3 and a half days to put tags on all those tracks. Then to get any useful data, you have to search that data. On top of hurdles such as that, the real trick for all that data is finding useful information. I did some math using stats from Google, and figured out that those 10 second clips will have about 20 words each. Now, it doesnt take much time to read 20 words, but doing that 34 million times? Thats a bit slower of a read. For you to be able to find anything useful in that data, you have to know what youre searching for, you have to know how to filter out what youre looking for from things that only really look like what youre looking for, and you have to know exactly what that all means. Knowing that since the release of Good Will Hunting that the Red Sox are 0-14 on Ben Afflecks birthday does you no good unless you know when Ben Afflecks birthday is, or who put the curse on him/the team and what you have to do to get it removed (Answers: Babe Ruth put the curse on them, and for it to be lifted the Mets have to have the best regular season record but still loose to the Astros in game 7). Itd be the same with this data. Unless you know what effect talking about McDonalds while posting a selfie when its raining has on McNugget sales, you cant make money off of it. And can you imagine how many times youll hear Bomb and Violent overthrow of the American government during during a Fox News broad cast (Im thinking 20 from the anchors alone)? Having a lot of data does mean you can make connections like this: for example, a study done by Cambridge and Microsoft shows that if your parents were separated before you were 21, you are more likely to like status that relate to relationships. While interesting as hell, its not exactly useful to most companies (If you want to check said study out, and you should, its here: pnas.org/content/early/2013/03/06/1218772110.full.pdf+html), but having data and doing something with it are two very different things.
Posted on: Thu, 29 May 2014 06:13:06 +0000
Trending Topics
Recently Viewed Topics
© 2015