ibm-watsonwatson-knowledge-studio

Use of the Negative mention class on WKS


While developing a custom machine learning model for the extraction of information on CVs using WKS, we started to wonder if the usage of the negative mention class feature for anottation only has effect on the model if the mention has a relation indicated with another one, the question arises because one can change the mention class from specific (the default class asigned on a first annotation task) to negative without the need of creating a relation, but we don't know if this way of using the negative mention class is useful at all.

We already know that the ML model learns from positive and negative examples (leaving a mention without annotation) will the use of the negative mention class feature improve the performance of the model further? we ask this because, clearly we don't know how WKS works in detail under the hood.


Solution

  • as you know, below answer was posted in the Communities Forum.

    Negative mention class can be used to annotate a mention of a lack of something. For example, "there is no police presence in the region" could have "police" annotated with entity type ORGANIZATION but mention class NEG. Another option however is not to use NEG and instead make one or more entity types for wording indicative of negation, doubt, or anything else other than certainty, and a relation type for connecting them to what they pertain to. So could have no/NEGATION appliesTo police/ORGANIZATION or some such.