Why use rules?

Rules are a useful approach in one or more of the following situations:

• The application you are developing has a requirement to analyze fixed expressions that include thousands or even millions of variants when it would be extremely difficult to provide enough data to learn a machine model. These kinds of expressions include numbers, monetary amounts, dates, and addresses, for example. It is hard for systems to learn models when the data is so diverse. Moreover, it is usually not necessary, as rules to analyze these expressions are not difficult to write because their formats are very structured. For both of these reasons, a rule-based approach is a simpler solution for recognizing fixed expressions.
• Very little training data is available for the application, and creating new data would be expensive. For example, annotating new data might require very specialized expertise. Although there are now techniques (such as few-shot or zero-shot learning) that can...