I am making a natural language query to Watson Discovery using NodeJS SDK, I am observing that sometimes its returning different number of passages/results for the same query and same configuration with steady document count. Even the scores returned for the passages are different when the passages are repeating in the subsequent result sets. I understand that Watson Discovery scoring mechanism would change when documents are getting added continuously. But we are testing on fixed set of documents where enrichment is done on test field for entities. I have passages and highlight fields as parameters to the query API of Watson Discovery.
Please see the below log for same natural language query:issue with outlook configuration
[{"documentId":"","documentType":"","title":"no server location in outlook (PCSEP79)","responseText":"Client sent e-mail that the issue with Outlook has been resolved.","sourceField":"text","score":16.041687463268694,"url":""}]
[{"documentId":"","documentType":"","title":"no server location in outlook (PCSEP79)","responseText":"Client sent e-mail that the issue with Outlook has been resolved.","sourceField":"text","score":1,"url":""},{"documentId":"","documentType":"","title":"Configuring the Outlook Client","responseText":"Launch Microsoft Outlook by clicking on the desktop icon or clicking Start > All Programs > Microsoft Office > Microsoft Outlook 2010 . Click Next on the Startup screen. Click Next on the configure an email account window.","sourceField":"text","score":0,"url":""}]
Kindly ignore discrepancy in score field as I am normalizing it in some scenarios-but question here is how Watson Discovery is returning different result sets for the same query and same data set. Is it its default behavior or something is wrong in my client code? Help of any kind would be greatly appreciated.
