AI and Machine Learning for On-Device Development

Chapter 5. Text Processing Apps with ML Kit on Android

Perhaps the two largest fields in machine learning are computer vision and natural language processing. In Chapter 4, you learned about some common computer vision scenarios with models that were already defined for you in ML Kit. In this chapter, you’ll explore some natural language processing ones, including how to recognize text from digital ink, perform smart replies to messages, and extract entities such as addresses from text. These are all off-the-shelf models for these specific scenarios. If you want to create apps that use other natural language processing models, such as text classification, you’ll have to create your own models using TensorFlow Lite, and then implement them on mobile. We’ll explore that in later chapters.

Entity Extraction

When given large amounts of text, extracting important information from it can be a difficult task. Often information that follows a particular structure, such as an address, might be predictable for one country but work very differently in another, so having a rules-based approach to acquire the information can lead to a lot of coding.

For example, consider Figure 5-1, where I’ve sent a message to my friend Nizhoni with some details in it. As a human we can extract valuable information about this, such as “tomorrow at 5 PM,” understanding that it’s a date and time. But writing code to do that can be really difficult. It’s hard enough trying to write code to understand formatted ...