Stopping false positive evidence being used for facts and anchors using forced preclusion of labels

In traditional classification we can often return false positives. That is, when looking for a phrase in the content, the phrase has indeed been found, but it is actually part of a LARGER phrase - and actually, the larger phrase shouldn’t be considered as evidence for the concept in the model.

For example, consider the concept “President”. When looking for evidence for this concept, we will, of course, be looking for the word “President” - and perhaps some other labels, such as “Barak Obama”, etc. However, if we had content that also spoke about the Vice President, then we have a problem, as our concept “President” will match on that phrase. But that, for us, is a false positive. it is not the President that was being spoken about, it was the vice President.

So, how do we stop that from happening? We use preclusions! Semaphore actually computes all of these preclusions it can find in the model’s labels. What it can’t do, is do the same for the content. It cannot automatically add at publish-time all of the false positive phrases to the rules that are going to occur in the content.

To do that, we must force the issue - by adding in “forced preclusions” to the model - so, for example, if your model had NOT contained a “Vice President” concept, then we could add a label to the President concept of “Vice President” as a special forced preclusion. Special preclusions rules WILL now be generated by Publisher, and that means that when Semaphore come across the larger phrase “vice president” in the content, it will make it impossible for “President” to fire as it will already have fired for the longer phrase “Vice President” which will have been scored with zero!

What does this have to do with FACTS? Well, the exact same issues occur with FACTS - both when trying to find the correct fact to extract, and also when trying to find the correct anchor to use. Both can be victims of false positives cropping up unexpectedly in content. However, we can follow the exact same method to stop that - by adding “forced preclusions” to concept facts / anchors. FACTS actually has a reserved alternative label type “has forced preclusion” for Concept Fact / Concept Anchor / Wildcard Fact / Wildcard Anchor types, which you should always use.

This has been modelled below, in FACTS:

What if our concepts and their labels are modelled in a taxonomy, and I am using a Taxonomy fact - what can I do then?

You can simply use the “has forced preclusion” alternative label on any of your concepts that you have created in taxonomies directly in the model or linked in form other models. See below: