LogosLink User's Manual · LogosLink version 2.0.0

Context Coverage Analytics (Agency Model)

Context Coverage analytics shows the degree of overlap between elements in the agency model and elements in the embedded context.

This is useful to evaluate how much of the context is referenced by the agency model, and how much of the agency model references the context.

Parameters

β. This allows you to weight precision and recall differently. Typical values are between 0.5 and 2.0. See the Details section below for more information.

Results

Results are given as some overall data and a tree table.

Overall data

Overall. This is the degree of coverage (overlap) between the agency model and the embedded context.
Overall weighted. This is the degree of coverage (overlap) between the agency model and the embedded context, but applying the β parameter to weight precision and recall differently. Please see the Details section below for more information.

An overall coverage of 0% indicates that the agency model contains no elements linked to the embedded context. An overall coverage of 100% indicates that the agency model and the embedded context fully overlap, that is, all the applicable elements in the agency model are linked to the context, and all elements in the context are linked to elements in the agency model.

Tree table

The tree table shows a hierarchy of agency model and context element kinds, with some details for each one.

Data in the tree table is aggregated hierarchically. For example, the data for "Speaker" and "Response Part" are added up and shown under "Agency Model Element".

For each element kind, the following data are shown:

Element kind. This indicates the kind of agency model or context element for which data is provided.
Cover count. This shows how many elements of the kind are linked to the other model. For agency model elements, this indicates how many of them are linked to context elements; for context elements, it indicates how many of them are linked to agency model elements.
Total count. This shows how many elements of the kind exist.
Coverage. This shows the percentage of the cover count in relation to the total count.

The Coverage value for the "Agency Model Element" entry in the tree table indicates how much of the agency model references the embedded context. Similarly, the Coverage value for the "Context Element" entry indicate how much of the context is referenced by the agency model.

Details

This analytics employs a precision and recall approach. Precision is interpreted as how many agency model elements are linked to context elements (i.e. how "precise" the agency model is in capturing its context), whereas recall is intepreted as how many context elements are linked to agency model elements (i.e. how much context is "recalled" by the agency model).

Similarly, the overall and weighted overall coverage measures are calculated as F₁ and F_β scores , respectively. Values of β between 0 and 1 make precision more important than recall in calculating the weighted overall coverage. Contrarily, values of β above 1 make recall more important than precision.