For the Q&A API, what algorithm does Watson use to determine the confidence percentage for each answer?
For the Q&A API, what algorithm does Watson use to determine the confidence percentage for each answer? I'm just wondering where the numbers come from. Thanks.
The algorithm details are considered confidential. Although I am not a subject matter expert in this area, I was able to track down some Watson architectural information from Artificial Intelligence magazine, which might be useful to you. See the following link: http://www.aaai.org/ojs/index.php/aimagazine/article/view/2303
Good question. Watson uses machine learning models to learn and predict confidence for each answer. These models leverage features generated by various algorithms and scoring techniques. Hope this helps.
I have looked into this, and the confidence does not seem fixed to the answer really. It seems that the confidence is related to the source though. In other words it is confident that the information is true by the credibility of the source. Whether or not it is related to the question has not been substantiated. @Lakshmikrishnamurthy, do you know if the algorithms have any names, ie mathematical algorithms etc. The OP asked what algorithm.