On January 11, 2016 PDT, we will release updates to two of our APIs, Topic Tagging and Text Classification.
Topic Tagging helps developers by automatically detecting the topics for any given text or string.
Text Classification can help sort a collection of unstructured data into predefined categories.
Prior to this update, Topic Tagging returned a list of topics where each topic had a number signifying the number of times that topic was found related in the text. This number was a bit hard to work with as it supposedly had no hard maximum value. After the update, a normalized value between 0 and 1 will be returned along with each topic instead.
Text Classification was another API that prior to this update, returned a score of how related it is to each category that a particular text was sorted into. This was another place where there was supposedly no hard maximum score. This will also change. A normalized value between 0 and 1 will be returned along with each category for a given text after the update.
A normalized value between 0 and 1 will make it easier for developers to test and set a threshold that fits with the data they are working with. Without an absolute maximum value like 1, it is hard to understand how confident a suggestion is.
This update will require developers to change their code BEFORE the January 11th update if their code depends on the values returned. To make things go smoother, we are providing a way for current developers to continue getting the current score. All you have to do is pass an extra parameter called legacy and set it to 1. (Please note this will eventually be deprecated also.)
If you have any issues with updating your code, please feel free to contact us at [email protected].