Method and apparatus for question-answering using a database consist of query vectors

1. A search method performed by a server, comprising:

receiving, using the server, a user question vector for a user question from a user terminal remote from the server;

selecting, using the server, a plurality of similar question candidates based on a first similarity analysis model determining a first similarity to the user question vector;

transmitting, using the server, the plurality of similar question candidates without answer information to the user terminal;

receiving, using the server, a selected question candidate selected by the user terminal from the plurality of similar question candidates from the user terminal, wherein the selected question candidate from the plurality of similar question candidates is selected by the user terminal based on a second similarity analysis model determining a second similarity to the user question performed at the user terminal;

generating, using the server, an answer to the selected question candidate selected by the user terminal as the answer to the user question; and

transmitting, using the server, the answer to the user question to the user terminal,

wherein the first similarity analysis model and the second similarity analysis model are different analysis models,

wherein the second similarity analysis model comprises paraphrase model, which is a neural network model learned using first learning data and second learning data,

wherein the first learning data composed of questions configured to have similar words and structures to have similar sentence formats and similar meanings,

wherein the second learning data composed of questions configured to have similar words and structures to have similar sentence forms, but have different meanings.

2. The search method of claim 1 , wherein

questions and question vectors for the questions are stored in association with each other in a database of the server.

3. The search method of claim 2 , wherein

the selecting of the plurality of similar question candidates based on the first similarity to the user question vector includes:

selecting questions associated with question vectors whose similarities to the user question vector are higher than a preset reference and which are stored in the database of the server as the plurality of similar question candidates.

4. The search method of claim 1 , wherein

the selecting of the plurality of similar question candidates based on the first similarity to the user question vector includes:

selecting the similar question candidates based on the first similarity analysis model, wherein the first similarity analysis model analyzes a distance in a Euclidean space between the user question vector.

5. The search method of claim 1 , wherein

the first learning data is composed of a first pair of questions and a label indicating that the first pair of questions are similar to each other, and

the second learning data is composed of a second pair of questions and a label indicating that the second pair of questions are dissimilar to each other.

6. The search method of claim 1 , wherein

the selecting of the plurality of similar question candidates based on the first similarity to the user question vector includes:

determining a similarity ranking of candidate questions belonging to the similar question candidates based on a similarity analysis result to the user question vector; and

selecting a predetermined number of candidate questions as the plurality similarity questions according to the similarity ranking.

7. The search method of claim 6 , wherein the selected question candidate from the plurality of similar question candidates is selected based in part on the ranking of the plurality of similar question candidates.

8. The search method of claim 1 , wherein the user question vector for the user question is generated at the user terminal.

9. A search server comprising:

a processor; and

a non-transitory memory configured to store instructions executed by the processor,

wherein the processor executes the instructions to:

receive a user question vector for a user question from a user terminal remote from the server,

select a plurality of similar question candidates based on a first similarity analysis model determining a first similarity to the user question vector,

transmit the plurality of similar question candidates without answer information to the user terminal,

receive a selected question candidate selected by the user terminal from the plurality of similar question candidates from the user terminal, wherein the selected question candidate from the plurality of similar question candidates is selected by the user terminal based on a second similarity analysis model determining a second similarity to the user question performed at the user terminal;

generate an answer to the selected question candidate selected by the user terminal as the answer to the user question, and

transmit the answer to the user question to the user terminal,

wherein the first similarity analysis model and the second similarity analysis model are different analysis models,

wherein the second similarity analysis model comprises paraphrase model, which is a neural network model learned using first learning data and second learning data,

wherein the first learning data composed of questions configured to have similar words and structures to have similar sentence formats and similar meanings,

wherein the second learning data composed of questions configured to have similar words and structures to have similar sentence forms, but have different meanings.

10. A non-transitory computer readable medium in which a computer program executed by a server is recorded, the computer program comprising:

receiving a user question vector for a user question from a user terminal remote from the server;

selecting a plurality of similar question candidates based on a first similarity analysis model determining a first similarity to the user question vector;

transmitting the plurality of similar question candidates without answer information to the user terminal;

receiving a selected question candidate selected by the user terminal from the plurality of similar question candidates from the user terminal, wherein the selected question candidate from the plurality of similar question candidates is selected by the user terminal based on a second similarity analysis model determining a second similarity to the user question performed at the user terminal;

generating an answer to the selected question candidate selected by the user terminal as the answer to the user question; and

transmitting the answer to the user question to the user terminal,

wherein the first similarity analysis model and the second similarity analysis model are different analysis models,

wherein the second similarity analysis model comprises paraphrase model, which is a neural network model learned using first learning data and second learning data,

wherein the first learning data composed of questions configured to have similar words and structures to have similar sentence formats and similar meanings,

wherein the second learning data composed of questions configured to have similar words and structures to have similar sentence forms, but have different meanings.