Hello John,

Thank you for your post: it will allow us to shed light on some important issues.

Firstly, I want to make it absolutely clear that the underlying algorithms are the SAME, and are common to both methods of invocation (soap and rest).

The most notable difference between the two is that SOAP layers (both client and server side) don't 'like' much the '\r' character, and for the most part just drop it altogether.

So, the text "hello\r\nworld" arrives at our end as "hello\nworld" when using soap.

This does not affect the extraction process, but it accounts for the "extra word" in the detection:  we try not to include broken words in the prefix/suffix of the detection, and so we count a certain amount of characters back and forward, and then look for the nearest word-boudary. Obviously, any extra character can exclude a whole word in/out, when using this logic.

Secondly, about the relevance scores: I was not able to reproduce that, and got identical scores.

My guess is that your two requests were submitted at different times.

Our relevance algorithm is based on an indexed collection of documents, and this "reference-collection" is updated on a regular basis. It seems to me that your more-recent request arrived into the system after this "DB" was updated, hence getting different scores.

Could you please confirm this by re-submitting both (soap and rest) at near times ?

HTH,

Regards,

Meir