Semantic Modeling for Data

Chapter 10. Bad Application

Computers are useless. They can only give you answers.

Pablo Picasso

So far we have seen pitfalls mainly related to the development of a semantic model. In this chapter, we switch perspective and we look at a frequent mistake that happens when we apply such a model in an application. The mistake is that we assume that just because the model has been designed for the same domain or kind of data the application operates in, its semantics are directly applicable and beneficial to it. In reality, it can be that:

• The application’s semantic needs seem to be covered by the model’s elements, yet there are subtle but crucial differences between them that make the model useless or even harmful

• The application’s semantic needs are covered by the model’s elements, but the model contains additional elements that are not just redundant but actually harmful to the application

In what follows, we see how these two issues can arise in two common applications of semantic models, namely entity resolution and semantic relatedness calculation, and how we can tackle them in each case.

Bad Entity Resolution

Entity resolution is an information-extraction task that involves detecting mentions of entities within texts and mapping them to their corresponding entities in a given semantic model. For example, consider the following text from an IMDb review of the 1997 film Steel:

How’s this for diminishing returns? In BATMAN AND ROBIN, George Clooney battled Arnold Schwarzenegger. ...