What kind of information extraction method is effective?

Generally speaking, the information extraction method based on statistics is more effective.

In terms of the difficulty of method implementation, statistical methods need to provide the topological structure of statistical models and a large number of training sets, while semantic (understanding)-based methods need to build a relatively complete natural language processing system, and rule-based (knowledge engineering) methods need to obtain rules for information extraction through knowledge engineering according to the language characteristics of different fields. In contrast, the method at the same level is easier to implement.

On the other hand, statistical methods are also superior to other methods in portability (or adaptability). The method based on statistics can establish a model suitable for another field by changing the training set.

But the reality is that information extraction is often concentrated in a small range of one or several fields, so the portability requirement is not high, and the application of rule-based information extraction in specific fields is easier to realize, so it can be said that the rule-based information extraction method is more effective.