Documentation

Automatic analysis of large text corpora is a complex task. This complexity particularly concerns the question of time efficiency. Furthermore, efficient, flexible, and extensible textanalysis requires the continuous integration of every new text analysis tools. Since there are currently, in the area of NLP and especially in the application context of UIMA, only very few to no adequate frameworks for these purposes, which are not simultaneously outdated or can no longer be used for security reasons, this work will present a new approach to fill this gap.

A pipeline is a collection of components or Analysis Engines that can be executed. During an analysis process, the components in the pipeline are executed one after another annotating documents. Pipelines do not interact with the input data directly but build the structure for an NLP workflow.

Creating a pipeline with this web-interface can be done in the Builder. It is a three step form that guides you through building a pipeline either from scratch or using a template as the starting point.

Choosing a template as a starting point copies all predefined settings into a fresh pipeline.

In the second step pipeline specific properties like name, description, tags and settings can be edited. Only a name is required to proceed but adding a short description is recommend to serve as documentation and help others when sharing a pipeline.

Try adding a tag to the pipeline on the right. below.

No tags added yet. Enter a value in the text field that says Add a tag... and press enter.

This is the state of your pipeline so far. Since there no components yet, the pipeline doesn't really do anything. Let's add some components to the pipeline.

A pipeline must have at least one component to be created.

Components are the part of DUUI that actually do the processing and therefore offer the most settings. When creating a pipeline you can choose from a set of predefined components or create your own. Once added to the pipeline, a component can be edited by clicking the icon. This will open a drawer on the right, that allows for modification of a component.

Options are specific to the selected driver. Most of the time the default options are sufficient and modifications are only for special uses cases. Parameters are useful if the component requires settings that are not controlled by DUUI.

When editing a specific pipeline, clicking the icon clones the component's settings and prefills the creation form.

A process manages the flow of data and pipeline execution. Starting a process is possible on a pipeline page. On the process creation screen you are asked to select an input, output and optionally settings that influence the process behavior.

The document is the smallest unit in DUUI and holds the data to be processed. Additionally documents track metrics about performance, integrity and performed annotations. Documents are presented in a table that shows basic information about their current state. Clicking on a row opens a more detailed view that includes the duration of different stages in the process, annotations that have been made and a treemap of said annotations.