0

My apologies if the question is duplicated. We are facing an issue with the analysis of a profile using Watson Personality Insights API in Spanish. We have a demo we implemented using PI API version 2 and then we tested the results (exact same text) with the demo published on developer cloud(in spanish) and we found important differences on how the big five were calculated when the facet values were not that different. Is it possible that these differences are caused because of the API version? The issue that with our demo the big five values produced a kind of negative summary profile when the developercloud summary is kinder. We could send both result jsons. For example here is how the big five rated:

BigFive DeveloperCloud Demo V2

Openness 0.773834349 0.847273232

Conscientiousness 0.916616088 0.914907481

Extraversion 0.796331544 0.612606551

Agreeableness 0.17445636 0.096118648

Emotional range 0.036287447 0.01623536

thanks in advance!!

Sonia
  • 1

2 Answers2

0

So the API version would not make a difference, as that just governs the format of the API; the back-end models are the same for both v2 and v3 of the API.

So the jist of your question is that when you run the same text in your app, and in the demo you get different big5 results, while the facet values are about the same.

This might be easiest solved by you opening a support ticket so we can debug the issue together; if you'd rather not do that then can you provide a sample text? Typically it boils down to a difference in the way the text is parsed.

Another question; did you try making the request using curl? That would cut out any custom logic in your app and narrow down the problem.

Neil Boyette
  • 171
  • 2
  • thanks @Neil! We tested this further using CURL with different inputs and comparing with IBM Demo results (always calling version 3 API and Spanish language). We tested sending the message with line breaks (pasted and copied it) on IBM demo, then we removed the line breaks and tested it again. Then we tested using CURL passing a plaintext with line breaks (same text we used in the first test on IBM demo), then using a json input without line breaks and finally using a json input with linebreaks (replaced the linebreaks by \r\n)...> continue next comment.. – Sonia Jun 11 '17 at 13:34
  • When comparing results we noticed the results differ if the used linebreaks or not. We got the exact same results when using IBM demo without line breaks and CURL json input without linebreaks, also the exact same results when using CURL with plaintext and json input with line breaks, but none of matched the IBM demo output with linebreaks. What we need to know is which are the correct results, the ones with line breaks or without them. And it would be interesting to know why IBM demo using linebreaks don’t macht any other results by far.. – Sonia Jun 11 '17 at 13:35
  • For example, for extraversion we got these results.. IBM demo with linebreaks: 0.7394869566810001 IBM demo without linebreaks: 0.552633041926584 CURL plain text with linebreaks: 0.5341493822131751 CURL json without linebreaks: 0.552633041926584 CURL json with linebreaks: 0.5341493822131751 – Sonia Jun 11 '17 at 13:35
0

thanks Neil for you answer! We tested the text using CURL and we noticed that the results didn't change by the service version used but instead by how the text was sent. If we called the service using curl passing a plain text input(formatted in UTF-8 with line breaks) it returned the same results for version2 and version 3 and also matched the ones from our demo. If we called the service using curl passing json input WITHOUT line breaks it returned the same values as well. But if we called the service passing the json input WITH line breaks then the results changed and almost matched those shown by ibm demo. My question here is which are the correct results? The ones shown when the text is sent as a plain text input(with line breaks) or when the text is sent as json input(with line breaks)? Is there any technical guideline besides the one shown in developercloud on how the text should be parsed to use this service? Thanks again!

Sonia
  • 1