Building Machine Learning Powered Applications

Chapter 7. Using Classifiers for Writing Recommendations

The best way to make progress in ML is through repeatedly following the iterative loop depicted in Figure 7-1, which we saw in the introduction to Part III. Start by establishing a modeling hypothesis, iterate on a modeling pipeline, and perform detailed error analysis to inform your next hypothesis.

Figure 7-1. The ML loop

The previous chapters described multiple steps in this loop. In Chapter 5, we covered how to train and score a model. In Chapter 6, we shared advice on how to build models faster and troubleshoot ML-related errors. This chapter closes an iteration of the loop by first showcasing methods to use trained classifiers to provide suggestions to users, then selecting a model to use for the ML Editor, and finally combining both to build a working ML Editor.

In “ML Editor Planning” we outlined our plan for the ML Editor, which consists of training a model that classifies questions into high- and low-score categories and use this trained model to guide users to write better questions. Let’s see how we can use such a model to provide writing advice to users.

Extracting Recommendations from Models

The goal of the ML Editor is to provide writing recommendations. Classifying a question as good or bad is a first step in this direction since it makes it possible to display the current quality of a question to a user. ...