Method and computer device for selecting a current context-specific response for a current user query

1. A method of selecting a current context-specific response for a current user query, the method executable by a server, the server being communicatively coupled to a database storing a plurality of vectors, each vector being associated with a respective conversation snippet, each conversation snippet including a respective context snippet and a respective response snippet, each vector having (i) a respective context vector indicative of features of the respective context snippet and (ii) a respective response vector indicative of features of the respective response snippet, the context vector and the response vector of each one of the plurality of vectors having been generated by a neural network,

the neural network having been trained:

based on a first context snippet and a first response snippet having occurred in a common conversation snippet and a second response snippet not having occurred with the first context snippet,

such that a positive-example vector similarity value between (i) a first response vector generated based on the first response snippet and (ii) a first context vector generated based on the first context snippet is superior to a negative-example vector similarity value between (i) a second response vector generated based on the second response snippet and (ii) the first context vector,

the method comprising:

receiving, by the server, a current context snippet comprising the current user query;

generating, by the neural network of the server, a current context vector based on the current context snippet, the current context vector being indicative of features of the current context snippet;

for each response snippet associated with a respective vector, generating, by the server, a respective ranking score being a sum of:

a respective first vector similarity value between (i) the current context vector and (ii) the respective context vector, each first vector similarity value being indicative of a linguistic similarity between (i) the current context snippet and (ii) the respective context snippet; and

a respective second vector similarity value between (i) the current context vector and (ii) the respective response vector, each second vector similarity value being indicative of a linguistic and a logical similarity between (i) the current context snippet and (ii) the respective response snippet; and

selecting, by the server, the current context-specific response amongst the response snippets based on the respective ranking scores.

2. The method of claim 1 , wherein the method further comprises ranking, by the server, the response snippets based on the respective ranking scores, and wherein the selecting the current content-specific response comprises selecting, by the server, a top ranked response snippet as the current context-specific response for the current user query.

3. The method of claim 1 , wherein the neural network comprises:

a context sub-network having generated the first context vector based on the first context snippet; and

a response sub-network having generated (i) the first response vector generated based on the first response snippet and (ii) the second response vector generated based on the second response snippet.

4. The method of claim 3 , wherein the generating the current context vector based on the current context snippet is executed by the context sub-network.

5. The method of claim 1 , wherein the neural network has been trained such that the positive-example vector similarity value is superior by at least a pre-determined threshold value over the negative-example vector similarity value.

6. The method of claim 1 , wherein the neural network has been trained to generate the first context vector and the first and second response vectors so as to maximize a difference between the positive-example vector similarity value and the negative-example vector similarity value.

7. The method of claim 1 , wherein the positive-example vector similarity value is indicative of a vector distance between (i) the first response vector and (ii) the first context vector, and wherein the negative-example vector similarity value is indicative of a vector distance between (i) the second response vector and (ii) the first context vector.

8. The method of claim 1 , wherein the positive-example vector similarity value is a scalar multiplication between (i) the first response vector and (ii) the first context vector, and wherein the negative-example vector similarity value is a scalar multiplication between (i) the second response vector and (ii) the first context vector.

9. The method of claim 1 , wherein a given first vector similarity value is a scalar multiplication between (i) the current context vector and (ii) the respective context vector, and wherein a given second vector similarity value is a scalar multiplication between (i) the current context vector and (ii) the respective response vector.

10. The method of claim 1 , wherein the method further comprises selecting, by the server, the plurality of vectors based on the current context snippet amongst all vectors stored in the database.

11. The method of claim 1 , wherein a given ranking score is a weighted sum of the respective first vector similarity value and the respective second vector similarity value.

12. The method of claim 1 , wherein the context vector and the response vector of each one of the plurality of vectors, which have been generated by a neural network, have been concatenated into the respective one of the plurality of vectors.

13. A computer device for selecting a current context-specific response for a current user query, the computer device being communicatively coupled to a database storing a plurality of vectors, each vector being associated with a respective conversation snippet, each conversation snippet including a respective context snippet and a respective response snippet, each vector having (i) a respective context vector indicative of features of the respective context snippet and (ii) a respective response vector indicative of features of the respective response snippet, the context vector and the response vector of each one of the plurality of vectors having been generated by a neural network,

the neural network having been trained:

based on a first context snippet and a first response snippet having occurred in a common conversation snippet and a second response snippet not having occurred with the first context snippet,

such that a positive-example vector similarity value between (i) a first response vector generated based on the first response snippet and (ii) a first context vector generated based on the first context snippet is superior to a negative-example vector similarity value between (i) a second response vector generated based on the second response snippet and (ii) the first context vector,

the computer device being configured to:

receive a current context snippet comprising the current user query;

generate, by the neural network, a current context vector based on the current context snippet, the current context vector being indicative of features of the current content snippet;

for each response snippet associated with a respective vector, generate a respective ranking score being a sum of:

a respective first vector similarity value between (i) the current context vector and (ii) the respective context vector, each first vector similarity value being indicative of a linguistic similarity between (i) the current context snippet and (ii) the respective context snippet; and

a respective second vector similarity value between (i) the current context vector and (ii) the respective response vector, each second vector similarity value being indicative of a linguistic and a logical similarity between (i) the current context snippet and (ii) the respective response snippet; and

select the current context-specific response amongst the response snippets based on the respective ranking scores.

14. The computer device of claim 13 , wherein the computer device is further configured to rank the response snippets based on the respective ranking scores, and wherein the computer device is configured to select the current content-specific response comprises the computer device being configured to select a top ranked response snippet as the current context-specific response for the current user query.

15. The computer device of claim 13 , wherein the neural network comprises:

a context sub-network having generated the first context vector based on the first context snippet; and

a response sub-network having generated (i) the first response vector generated based on the first response snippet and (ii) the second response vector generated based on the second response snippet.

16. The computer device of claim 15 , wherein the computer device is configured to generate the current context vector based on the current context snippet by employing the context sub-network.

17. The computer device of claim 13 , wherein the neural network has been trained such that the positive-example vector similarity value is superior by at least a pre-determined threshold value over the negative-example vector similarity value.

18. The computer device of claim 13 , wherein the positive-example vector similarity value is indicative of a vector distance between (i) the first response vector and (ii) the first context vector, and wherein the negative-example vector similarity value is indicative of a vector distance between (i) the second response vector and (ii) the first context vector.

19. The computer device of claim 13 , wherein the positive-example vector similarity value is a scalar multiplication between (i) the first response vector and (ii) the first context vector, and wherein the negative-example vector similarity value is a scalar multiplication between (i) the second response vector and (ii) the first context vector.

20. The computer device of claim 13 , wherein a given first vector similarity value is a scalar multiplication between (i) the current context vector and (ii) the respective context vector, and wherein a given second vector similarity value is a scalar multiplication between (i) the current context vector and (ii) the respective response vector.