Model is not able to identify new questions form user that are not in Alternate Questions

Hi there,

We have an on premise which uses version 7.2. We are using Knowledge Graph to train some FAQs. But we are facing issues identifying new phrases from user that are not in the added alternate questions.

We are using the “Tags” and path level synonyms to train each alternate questions. The user query does containing the exact tag as in the trained alternate question. But still the model fails to identify the question.

Please see the below example:

Alternate question: Do I need to migrate to <application_name> version 2?

Tags used:

  • migrate to <application_name> version 2 (mandatory tag)
  • <application_name> version 2 (mandatory tag)

Query that the user asked:
why should I migrate to <application_name> version 2

As you can see the same mandatory tag is present in the query that the user is asked and the query itself is similar to the one that we have trained.

There is another closely related query that is in the knowledge graph is “How can I migrate to <application_name> version 2?”. These are 2 separate questions and have different responses trained in the knowledge graph.

Is it because the model is getting confused of the two? Whenever we are asking queries related to these 2 questions, the FAQ score is around 70% only and the confidence is marking as “low”.

Please provide us some insight into this issue.

Ordinarily in KG the question words, “how”, “why”, “what” are consider as stop words and therefore not part of the matching.
If you have two questions that differ only on which question word is being used then you can add them as an explicit mandatory tag.