Identifying entities through annotators

Let's return to the first style of identifying an entity that you care about. In the first approach, we showed you in Recognizing entities, we went to the intent examples, and highlighted the entity that we cared about in that example. We refer to this as the annotator approach to entity recognition. Annotators are a much more powerful way of identifying the entities. Rather than setting a rule, such as the word hot is a temperature entity, the annotator approach builds a machine learning algorithm that takes in the context of the surrounding words in the sentence. In doing so, it calculates the probability that hot is referring to a temperature. In a sentence such as The water is hot, it makes sense to recognize hot as being a temperature. But in another sentence, such as The market is hot, it should be taken to be the relative value people place on a particular product.

As your conversational agent gets more sophisticated, you will likely want to make more use of the annotator approach to entity recognition, as doing so will make its interpretation more robust. However, keep in mind that, like most machine learning problems, you have to teach Watson through examples—generally with several examples. If you're going to use this approach, you should find several examples in your intent training set in which you make use of the different entity types you care about, and label them accordingly.

So, in this example, we have added a few examples of the #Funds-Transfer intent, and, within that we have annotated, (as indicated by their highlights in the following example) a number of @Account entities: