KR20250045973A - System of providing natural language response to user queries - Google Patents

Abstract

The system for providing a natural language response to a user's query of the present invention includes an authentication key generation server which is connected to an internal network and generates and provides a predefined authentication key to a user terminal located on the same internal network, a database which stores text-based data by field, and a query response server which divides the text-based data by field stored in the database into string units, extracts features from sentences or words in the divided strings, regenerates feature information by field through the extracted features, transmits a string including feature information corresponding to a query input from a user terminal to a natural language processing server, and provides a natural language processed string received from the natural language processing server to the corresponding user terminal. The query response server is characterized in that it authenticates an authentication key provided to the user terminal from the authentication key generation server and provides a natural language response according to the query of the corresponding user terminal only when the authentication key is valid.

Description

{SYSTEM OF PROVIDING NATURAL LANGUAGE RESPONSE TO USER QUERIES}

The present invention relates to a system for providing a natural language response to a user query, which provides a response to a query input from a user terminal in the form of natural language.

Recently, with the advancement of AI technologies such as ChatGPT, technologies have emerged that perform web searches for queries requested by users and provide answers accordingly.

This AI question-answering technology goes through a series of processes to identify the questioner's intent, extract data appropriate to the intent from an already built web environment, and reprocess the extracted data into the form of an answer desired by the user.

Meanwhile, since such AI-based question-answering technology requires server-level resources for web data search, similarity analysis, and natural language processing, and it costs a lot of money to build a server that performs the corresponding functions, it is difficult for general users to build it on their own, and the service has been provided in the form of users using a question-answering service in which the corresponding environment has already been built and paying only for the amount of usage. The technology related to such natural language question-answering is disclosed in Republic of Korea Patent Publication No. 10-0452024 (Application date: 2001.01. 08., Publication date: 2004.10. 08., hereinafter referred to as “prior art”).

However, since the prior art performs the entire process of selecting, searching, and responding to meaningful data in a web environment in response to a user's query, the cost incurred in response to the user's query may be set high, and since it is not possible to verify whether the user requesting the query is valid, unnecessary costs may be incurred due to query requests by unauthorized users.

In addition, since the conventional technology provides responses based on search results in the form of short answers, it is difficult to deliver accurate responses depending on the user's vocabulary, and there is a problem that if a lot of unnecessary noise occurs during the data selection process in web-based data, there is a high possibility of deriving non-professional responses.

Furthermore, in the independent/closed web service environment established in public institutions, etc., the authority and connection environment for accessing the API (Application Programming Interface) of an external server are not provided, but since the conventional technology cannot be applied in the above-described environment, there was a problem in that external Open AI services such as natural language processing could not be provided.

The present invention aims to provide a natural language response system for user queries that can verify whether a user's query request is valid, minimize the cost incurred in processing a natural language response to the verified user query, provide a highly transferable query response, and provide an external service even in an independent/closed web service environment, in order to solve the above-described problem.

According to one embodiment of the present invention, a system for providing a natural language response to a user query includes: an authentication key generation server which is connected to an internal network and generates and provides a predefined authentication key to a user terminal located on the same internal network; a database in which text-based data by field is stored; a query response server which divides the text-based data by field stored in the database into string units, extracts features from sentences or words in the divided strings, regenerates feature information by field through the extracted features, transmits a string including feature information corresponding to a query input from a user terminal to a natural language processing server, and provides a natural language processed string received from the natural language processing server to the corresponding user terminal; wherein the query response server authenticates an authentication key provided to the user terminal from the authentication key generation server, and provides a natural language response according to the query of the corresponding user terminal only when the authentication key is valid.

Here, the authentication key generation server is characterized by generating an authentication key including the number of natural language response requests and the natural language response request validity period and providing the authentication key to the user terminal.

In addition, the query response server includes a communication unit that communicates with the database, the user terminal, and the natural language processing system; an authentication unit that authenticates an authentication key of the user terminal provided from the authentication key generation server; a collection unit that collects text-based data by field stored in the database; an extraction unit that separates the text-based data by field collected by the collection unit into string units and generates feature information from sentences or words in the separated strings; and a comparison unit that compares a query string received by the communication unit from a user terminal whose authentication key has been confirmed to be valid through the authentication unit with the feature information generated by the extraction unit, and selects a string including feature information with the highest degree of matching; and the comparison unit is characterized in that it extracts features in the query string received from the user terminal by the communication unit through the extraction unit, and selects feature information including the feature information with the highest degree of matching with the extracted feature.

And, the extraction unit is characterized in that, only when the collection unit collects text-based data for any field added or changed from the database, the text-based data for the field is re-separated into strings, and features are extracted from sentences or words within the separated strings to regenerate feature information for each field.

In addition, the above extraction unit is characterized by generating feature information from the first value for any word or sentence in each text-based data through the following [Formula 1] .

[Formula 1]

In addition, the above extraction unit is characterized by generating feature information from the second value for any word or sentence in the entire text-based data through the following [Formula 2] .

[Formula 2]

At this time, the extraction unit is characterized by generating feature information in the form of a real number vector for an arbitrary word or sentence by performing a product operation of the first value and the second value.

Meanwhile, the comparison unit is characterized by comparing features corresponding to words or sentences in a query string received from the user terminal with feature information for each word or sentence generated in advance, identifying feature information with a high degree of similarity, and assigning a real number vector value of the identified feature information to the feature in the query string.

In addition, the present invention is characterized in that the cosine similarity between the real number vector values of words or sentences in the query string and the real number vector values of the feature information of each word or sentence generated from the extraction unit is checked by the base comparison unit, and a string including a word or sentence with a high cosine similarity is selected as a string suitable as a response to the query string.

The system for providing a natural language response to a user query according to the present invention uses an authentication key issuance method of an authentication key generation server, and the query response server receives a query request from a user terminal, transmits the content to a natural language processing server, and provides a natural language processed response from the natural language processing server to the user terminal only for a user terminal to which a valid authentication key has been issued, thereby enabling a user terminal located in an independent/closed web service environment to use a natural language processing service on an external network based on the authentication key.

In addition, the system for providing a natural language response to a user query according to the present invention selects a character string within text-based data of the same field that is most suitable for a query received from a user terminal, transmits only the selected character string to a natural language processing server, and provides a naturally processed response to the user terminal, thereby having the effect of providing a natural language response of the same level while minimizing the cost incurred compared to a natural language response processing method for the entire query.

FIG. 1 is a schematic diagram of a system for providing natural language responses to user queries according to a preferred embodiment of the present invention.
FIGS. 2 and 3 visualize a method for providing a natural language response by each configuration of a question response server in a system for providing a natural language response to a user query according to a preferred embodiment of the present invention.

Hereinafter, a preferred embodiment of a natural language response system for a user inquiry of the present invention will be described in detail with reference to the attached drawings.

The same reference numerals presented in each drawing represent the same members. In addition, specific structural and functional descriptions of the embodiments of the present invention are merely illustrative for the purpose of explaining the embodiments according to the present invention, and unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those of ordinary skill in the art to which the present invention belongs. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning they have in the context of the related art, and are preferably not interpreted in an ideal or excessively formal meaning unless explicitly defined herein.

FIG. 1 is a schematic diagram of a system for providing a natural language response to a user query according to a preferred embodiment of the present invention, and FIGS. 2 and 3 are visual diagrams of a method for providing a natural language response by each configuration of a query response server in a system for providing a natural language response to a user query according to a preferred embodiment of the present invention.

Referring to FIGS. 1 to 3, a system for providing a natural language response to a user query according to a preferred embodiment of the present invention may include a query response server (100), an authentication key generation server (200), and a database (300).

Hereinafter, each configuration and operation will be described only for the purpose of processing a query requested from a user terminal (400) that is located in an independent/closed web service environment and cannot make a direct request to a natural language processing server (500) located in an external network through a query response server (100) according to a preferred embodiment of the present invention, performing natural language processing on the natural language processing server (500), and then providing the query back to the user terminal (400). However, the present invention is not limited thereto, and it should be understood that the present invention can be applied to various environments in addition to an independent/closed web service environment. In addition, for the convenience of explanation, the network in an independent/closed environment and the external service network that is not directly reached from the independent/closed environment will be described by replacing them with ‘internal network’ and ‘external network’, respectively.

Before explaining the question response server (100) according to a preferred embodiment of the present invention, the authentication key generation server (200) and database (300), which are other components of the system for providing natural language responses to user queries according to a preferred embodiment of the present invention, will first be explained.

The authentication key generation server (200) is connected to an internal network and can generate and provide a predefined authentication key to a user terminal located on the same internal network. Here, the authentication key generated by the authentication key generation server (200) can include the internal network name (=organization name) where the user terminal (400) is located, the name of the service to be used, the IP address of the user terminal (400), a unique token and the number of natural language response requests, and the natural language response request validity period, so that only an authorized user terminal (400) on the internal network can perform natural language processing on behalf of a query response server (100) that mediates a natural language processing server (500) on an external network through API access for the request content (=query) of the user terminal (400), so that the user terminal (400) can receive query requests and responses through the query response server (100) for a predefined number of times and period without issuing an authentication key for each query request.

The database (300) can store text-based data by field. Here, the text-based data is provided in the form of a document or a text file, and in some cases, it can be a document form that mixes text and images. Examples of text-based data include a user manual for a specific product, an encyclopedia, and the like, and documents produced for the purpose of providing information can be included. In addition, the text-based data can be stored in a form grouped by various fields, so that the collection unit (130) of the question-and-answer server (100) described below can classify and collect pre-classified text-based data by field.

A query response server (100) according to a preferred embodiment of the present invention verifies whether an authentication key of a user terminal (400) that has received an authentication key from a generation server (200) upon authentication is valid, receives a query input from a user terminal (400) that has been confirmed to be valid, compares features in the received query with features pre-configured from a plurality of text-based data in a database (300), selects a character string including features similar to the user query, transmits the selected character string to a natural language processing server (500) to provide natural language processed content to the user terminal (400), and may include a communication unit (110), an authentication unit (120), a collection unit (130), an extraction unit (140), and a comparison unit (150).

The communication unit (110) can communicate with the database (300), the user terminal (400), and the natural language processing server (500) through the communication network. Here, the communication unit (110) is connected to the database (300) and the user terminal (400) through the internal network, and is connected to the natural language processing server (500) through the external network. After processing a query requested from the user terminal (400) on the internal network, the communication unit provides the processed result of the natural language processing server (500) located on the external network, and can provide the final result of the natural language processing to the user terminal (400).

The authentication unit (120) can authenticate the authentication key of the user terminal (400) provided from the authentication key generation server (200). Here, the authentication unit (120) verifies whether the authentication key is a currently available authentication key by checking the number of request validity periods, request validity periods, etc. included in the authentication key generated from the authentication key generation server (200), and determines whether to provide a query request from the user terminal (400) and a response to the user terminal (400). In addition, the authentication unit (120) can perform encryption and decryption of the query content and the query response input from the user terminal (400) that provided the authentication key determined to be valid. In addition, the authentication unit (120) can additionally store information including the IP address of each user terminal (400), the ID of the user, etc., and determines whether to provide a query request and response from the user terminal (400) and the user based on the authentication result.

The collection unit (130) can collect text-based data that is pre-classified by field within the database (300) through the communication unit (110) described below. Here, the text-based data collected by the collection unit (130) is prepared in the form of a document or a text file, and in some cases, the collection unit (130) can extract and collect only text from a document containing text and images. In addition, the collection unit (130) can detect additions or changes to text-based data for a specific field within the database (300), and can newly collect the detected text-based data or replace it with previously collected text-based data.

The extraction unit (140) can separate the field-based text data collected from the collection unit (130) into string units and generate feature information from sentences or words within the separated strings. Here, the extraction unit (140) can group the separated strings and the feature information generated within each string unit into each text-based data unit, and the grouped string units and feature information can also be classified by field. In addition, when a query is input from a user terminal (400) through the communication unit (110) described below, the extraction unit (140) can extract features from the input query string. At this time, the features extracted from the query string can be sentences or words. Meanwhile, the field-specific feature information generated by the extraction unit (140) may be information that represents each word or sentence included in each field-specific text-based data or the entire field-specific text-based data in real-valued vector units according to frequency of occurrence and importance, and the extraction unit (140) may calculate a first value corresponding to the term frequency (TF) of any word or sentence in each field-specific text-based data through the following [Formula 1].

[Formula 1]

The extraction unit (140) can calculate a second value corresponding to the importance (IDF, Inverse Document Frequency) of any word or sentence in the entire text-based data by field through [Formula 2] below.

[Formula 2]

Thereafter, the extraction unit (140) performs a multiplication operation on the first value and the second value to generate feature information in the form of a real number vector for any word or sentence, so that the extraction unit (140) can express the word or sentence in each text-based data as a real number vector value through the generated feature information.

The comparison unit (150) can compare the features in the query string extracted from the extraction unit (140) with the feature information previously generated by the extraction unit (140), and select a string including feature information with a high degree of agreement and importance. Here, the comparison unit (150) assigns a real number vector value corresponding to each feature in the user query, that is, each word or sentence in the query string input from the user terminal (400), through the feature information in the form of a real number vector generated by the extraction unit (140) prior to receiving the query from the user terminal (400). In this process, the comparison unit (150) first checks text-based data having a large number of matching features, selects text-based data in a field corresponding to the query content, and checks the cosine similarity between the real vector values for words or sentences in the query string to which real vector values are assigned and the real vector values of the feature information of each word or sentence in the text-based data of the same field generated from the extraction unit (140), so that a string including a word or sentence with a high cosine similarity can be selected as a ‘KEY’ string suitable as a response to the query string. Thereafter, the ‘KEY’ string selected by the comparison unit (150) is transmitted to the natural language processing server (500) through the communication unit (110), and the natural language query response content received by the natural language processing server (500) through natural language processing is provided to the user terminal (400) again through the communication unit (110), so that the user can obtain a natural language response to the query.

That is, a system for providing a natural language response to a user query according to a preferred embodiment of the present invention comprises a query response server (100) that collects text-based data by field stored in a database (200), separates each text-based data into a string unit, generates feature information from words or sentences included in each string, and when a user terminal (400) that has received an authentication key from an authentication key generation server (200) located on the same internal network requests a query to the query response server (100), the query response server (100) checks whether the authentication key of the user terminal (300) is valid, and if valid, extracts features from words or sentences in a query string input from the user terminal (400), selects feature information with the highest similarity to the extracted features, and provides the feature information to the natural language processing server (500), and when the natural language processing is completed by the natural language processing server (500), the natural language processed string is provided to the user terminal (500), thereby enabling a natural language response to the user query to be provided.

Therefore, the system for providing a natural language response to a user query according to the present invention has the effect of allowing a user terminal located in an independent/closed web service environment to use a natural language processing service on the external network, as the query response server receives a query request from a user terminal, transmits the content to a natural language processing server, and provides a natural language processed response from the natural language processing server to a user terminal only for a user terminal to which a valid authentication key has been issued, through an authentication key issuance method of an authentication key generation server.

In addition, the system for providing a natural language response to a user query according to the present invention selects a character string within text-based data of the same field that is most suitable for a query received from a user terminal, transmits only the selected character string to a natural language processing server, and provides a naturally processed response to the user terminal, thereby having the effect of providing a natural language response of the same level while minimizing the cost incurred compared to a natural language response processing method for the entire query.

The above-described embodiments of the present invention have been disclosed for the purpose of illustration, and those skilled in the art with common knowledge of the present invention will be able to make various modifications, changes, and additions within the spirit and scope of the present invention, and such modifications, changes, and additions should be considered to fall within the scope of the following patent claims.

100: Query Response Server
110: Communications Department
120: Authentication Department
130: Collection Department
140: Extraction section
150: Comparison Department
200: Authentication key generation server
300: User terminal
400: Database
500: Natural Language Processing Server

Claims (9)

An authentication key generation server that is connected to an internal network and provides a predefined authentication key to a user terminal located on the same internal network;
A database that stores text-based data by field;
A query response server is included, which separates field-specific text-based data stored in the above database into string units, extracts features from sentences or words within the separated strings, regenerates field-specific feature information through the extracted features, transmits a string containing feature information corresponding to a query input from a user terminal to a natural language processing server, and provides a natural language-processed string received from the natural language processing server to the corresponding user terminal.
A system for providing a natural language response to a user's query, characterized in that the query response server authenticates an authentication key provided to a user terminal from the authentication key generation server and, only when the authentication key is valid, provides a natural language response according to the query of the user terminal.
In the first paragraph,
A system for providing a natural language response to a user query, characterized in that the authentication key generation server generates an authentication key including the number of natural language response requests and the natural language response request validity period and provides the authentication key to the user terminal.
In the second paragraph,
The above query response server,
A communication unit communicating with the above database, user terminal, and natural language processing system;
An authentication unit that authenticates the authentication key of a user terminal provided from the authentication key generation server;
A collection unit that collects field-specific text-based data stored in the above database;
An extraction unit that separates the field-specific text-based data collected by the above collection unit into strings and generates feature information from sentences or words within the separated strings; and
The above communication unit includes a comparison unit that compares a query string received from a user terminal whose authentication key has been confirmed to be valid through the authentication unit with the feature information generated by the extraction unit, and selects a string including the feature information with the highest degree of matching;
A system for providing a natural language response to a user query, characterized in that the comparison unit extracts features from a query string received from the user terminal through the communication unit through the extraction unit, and selects feature information including the feature information with the highest degree of consistency with the extracted features.
In the third paragraph,
A system for providing a natural language response to a user's query, characterized in that the extraction unit re-separates text-based data of the field into string units only when the collection unit collects text-based data for any field added or changed from the database, and extracts features from sentences or words within the separated strings to regenerate feature information by field.
In paragraph 4,
A system for providing a natural language response to a user query, characterized in that the above extraction unit generates feature information from a first value for any word or sentence in each text-based data through the following [Formula 1] .
[Formula 1]

In paragraph 5,
A system for providing a natural language response to a user query, characterized in that the above extraction unit generates feature information from a second value for any word or sentence in the entire text-based data through the following [Formula 2] .
[Formula 2]

In Article 6,
A system for providing a natural language response to a user query, characterized in that the above extraction unit performs a product operation of the first value and the second value to generate feature information in the form of a real number vector for an arbitrary word or sentence.
In Article 7,
A system for providing a natural language response to a user query, characterized in that the comparison unit compares the features corresponding to the words or sentences in the query string received from the user terminal with the feature information for each word or sentence generated in advance, identifies the feature information with a high degree of similarity, and assigns the real number vector value of the identified feature information to the feature in the query string.
In Article 8,
A system for providing a natural language response to a user query, characterized in that the above comparison unit checks the cosine similarity between the real number vector values for words or sentences in the query string and the real number vector values of the feature information of each word or sentence generated from the above extraction unit, and selects a string containing a word or sentence with a high cosine similarity as a string suitable as a response to the query string.