1. A method for delivering response to a user based on a recognized context‐
aware intent of the user, the method comprising:
‐ generating, by a processing unit [102], a present turn system‐response
based on a previous turn user‐response and a previous turn intent;
‐ receiving, by the processing unit [102] via a user interface [110], a
present turn user‐response;
‐ generating, by the processing unit [102] via an embeddings generator
unit [104], one or more embedding vectors for the present turn
system‐response, and one or more embedding vectors for the present
turn user‐response;
‐ generating, by the processing unit [102] via a concatenation unit [106],
a concatenated vector embedding based on the one or more
embedding vectors for the present turn system‐response and the one
or more embedding vectors for the present turn user‐response;
‐ generating, by the processing unit [102] via an intent recognition unit
[108], a present turn intent based on the concatenated vector
embedding and the previous turn intent; and
‐ delivering, by the processing unit [102] via the user interface [110], a
next turn system‐response based on the concatenated vector
embedding and the present turn intent.
2. The method as claimed in claim 1, wherein the intent recognition unit [108]
implements an intent recognition model, wherein the intent recognition
model is trained based on a combination of cross entropy losses, and
wherein the combination of cross entropy losses comprises a first
weighted loss and a second weighted loss.
22
3. The method as claimed in claim 1, wherein the generating, by the
processing unit [102] via the intent recognition unit [108], the present turn
intent based on the concatenated vector embedding and the previous turn
intent further comprises generating the present turn intent based on a first
weighted logits and a second weighted logits.
4. The method as claimed in claim 1, wherein prior to the generating, by the
processing unit [102] via the concatenation unit [106], the concatenated
vector embedding, the method comprises:
‐ performing, by the processing unit [102], a truncation process on the
present turn system‐response.
5. The method as claimed in claim 1, wherein prior to the delivering, by the
processing unit [102] via the user interface [110], the next turn system‐
response, the method comprises:
‐ triggering, by the processing unit [102], an idiom tool.
6. The method as claimed in claim 1, wherein the present turn system‐
response is a first turn system‐response and the previous turn intent is a
one‐hot encoded vector with all zero values.
7. A system for delivering response to a user based on a recognized context‐
aware intent of the user, the system comprising:
23
‐ a processing unit [102] configured to:
o generate a present turn system‐response based on a previous
turn user‐response and a previous turn intent;
o receive, via a user interface [110], a present turn user‐response;
o generate, via an embeddings generator unit [104], one or more
embedding vectors for the present turn system‐response, and
one or more embedding vectors for the present turn user‐
response;
o generate, via a concatenation unit [106], a concatenated vector
embedding based on the one or more embedding vectors for
the present turn system‐response and the one or more
embedding vectors for the present turn user‐response;
o generate, via an intent recognition unit [108], a present turn
intent based on the concatenated vector embedding and the
previous turn intent; and
o deliver, via the user interface [110], a next turn system‐
response based on the concatenated vector embedding and the
present turn intent.
8. The system as claimed in claim 7, wherein the intent recognition unit [108]
is configured to implement an intent recognition model, wherein the intent
recognition model is trained based on a combination of cross entropy
losses, and wherein the combination of cross entropy losses comprises a
first weighted loss and a second weighted loss.
24
9. The system as claimed in claim 7, wherein the processing unit [102] via the
intent recognition unit [108], is configured to generate the present turn
intent based on a first weighted logits and a second weighted logits.
10. The system as claimed in claim 7, wherein the processing unit [102], prior
to the generating, via a concatenation unit [106], a concatenated vector
embedding, is configured to:
‐ perform a truncation process on the present turn system‐response.