NLP Workbench Web Services Architecture

Diagram showing the suggested Natural Language Processing (NLP) Workbench Web Services architecture consisting of user interfaces, web service components, NLP applications, and offline components.

The diagram above shows the suggested architecture of the Natural Language Processing (NLP) Workbench Web Services. The architecture is described below.

User Interfaces

The NLP Workbench will provide access to two types of users—

  • Engineers will have access to tools and features needed to develop NLP applications and address specific domain needs.
  • End users will use the NLP applications developed by the engineers. For example, end users will use eMaRC Plus and ETHERExternal (Event-based Text-mining of Health Electronic Records system) interfaces to access the corresponding NLP web services developed by CDC and FDA engineers to process the unstructured pathology and safety data, respectively.

Web Service Components

In the web service messaging infrastructure, web service consumers (safety data, pathology, and general NLP service consumers) send requests consisting of JSON or XML messages to web service providers (safety data, pathology, general NLP, and data conversion service providers), which respond in kind. The web service messaging infrastructure interacts with the user interfaces and the NLP applications.

NLP Applications

NLP applications include extracting temporal and clinical information, summarizing this information, encoding it, and structuring it in the Common Data Model. Other NLP applications may be developed as well.

Offline Components

  • Rule-based models will be used to summarize the temporal and clinical information.
  • Hybrid approaches will be used to structure the encoded in the Common Data Model.
  • Trained machine learning models will be used to develop offline training for machine learning and training datasets.