Post the release of the latest OS by Google named Android 9 Pie, the company has provided insights into the A.I. based processes that explain the vastly improved algorithm in the OS that successfully creates actionable links for the phone numbers as well as addresses present in the text. Even though the concept isn’t something new, Google has managed to introduce a better version for the Android Linkify API which has been termed as Smart Linkify.
It is basically an API trained as TextClassifier that works as TensorFlow with the use of minuscule feedforward based neural network. This also includes a similar code for interface that comes with the Smart Text Selection by Oreo. This means that the AI can easily understand whether or not a series of words or numbers works as an address or phone number. This is similar to what Android Oreo had. However, this new Android doesn’t just make it easier for highlighting and copying associated text, it actually adds a link that is contextually relevant for the users to opt for immediate action as opposed to clicking it straight away as opposed to selection of this text manually.
That being said, this underlying process tends to be more complex than it seems. It goes way beyond just teaching A.I. the pattern for recognition of conversational ways with which the addresses, phone numbers, as well as other entities tend to be written out. Additionally, the AI also needs to make sure that the A.I. is being associated with appropriate action to be taken with text items discovered by it.
The A.I. has been trained to differentiate between the patterns of phone numbers of any shipment identification number. Google has included a series of training patterns to help the A.I. treat the numbers in their own way instead of intermixing them and a similar stance has been implemented for the words or phrases.
The A.I. works through 2 small neural networks made for feedforward that looks for the context of the surrounding words. The first feedforward tends to analyze every possible combination for different entries which are referred as candidates after which the same is scaled over the validity to come out with an actual option. Next, the AI removes candidates which tend to overlap way before 2nd neural network tends to take over the word to assign particular type for the analysis of similar potential candidates. Currently, Google is working over the further improvement of the A.I. that can make identification better.