KR102806164B1 - Method and apparatus of performing tagging process for recommending policy to individual based on natural language-processed policy data - Google Patents

Abstract

The present invention relates to a method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data performed by a server, comprising: (a) a step of receiving keywords extracted from a policy announcement through a preset policy data natural language processing learning model; (b) a step of selecting keywords corresponding to preset category items from among the keywords extracted according to step (a) by considering weights through a named entity recognition module, and generating an identifier indicating a category item corresponding to the selected keyword; and (c) a step of generating a classification result table by entering the selected keyword into a category item corresponding to the keyword based on the identifier, and in step (b), a method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, wherein the identifier is generated for a keyword related to at least one of the residence, type of business, income, age, gender, and number of children of an individual who is a target of policy recommendation from among the keywords extracted according to step (a).

Description

{METHOD AND APPARATUS OF PERFORMING TAGGING PROCESS FOR RECOMMENDING POLICY TO INDIVIDUAL BASED ON NATURAL LANGUAGE-PROCESSED POLICY DATA}

The present invention relates to a method and device for performing a tagging process for recommending policies to individuals based on policy data processed in a natural language, and more specifically, to tagging and classifying policy data processed in a natural language by a policy data natural language processing learning model for the purpose of recommending policies to individuals.

In the past, in order to support a policy implemented by the state, users who wanted to apply directly read the policy announcement and searched for the target policy. In this case, there was a problem in that unnecessary information was provided to users due to insufficient knowledge even though they were not eligible for support by the policy.

In particular, for individuals, the writing method and standard laws for policy announcements differ depending on the organization, department, and person in charge, making it difficult to determine whether the policy is for companies or individuals.

Even if the policy announcements issued by government agencies are refined by extracting data using natural language processing, text processing technology, and entity recognition technology, only about 7,000 characters of sentence components are extracted, so the need for technology to refine and digitize this arises.

The present invention is intended to solve the problems of the above-mentioned prior art, and has as its technical task to provide an automated tagging method for recommending policies to individuals through natural language processed policy data.

In addition, the present invention aims to facilitate the utilization of collected policy data for policy recommendation more conveniently than conventional methods by performing a process of tagging keywords as identifiers and classifying the results of natural language processing of policy data extracted from a policy announcement and categorizing them into various category items based on the keywords.

In addition, in the process of tagging and classifying policy data, if the policy is for an individual, the policy data that meets the conditions can be tagged and classified based on the weight given according to the preset algorithm.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood from the description below.

As a technical means for achieving the above-described technical task, according to an embodiment of the present invention, a method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data performed by a device comprises: (a) a step of receiving keywords extracted from a policy announcement through a preset policy data natural language processing learning model; (b) a step of selecting keywords corresponding to preset category items from among the keywords extracted according to step (a) by considering weights through a named entity recognition module, and generating an identifier indicating a category item corresponding to the selected keyword; and (c) a step of generating a classification result table by entering the selected keyword into a category item corresponding to the keyword based on the identifier, wherein in step (b), the identifier may be generated for a keyword related to at least one of the residence, business type, income, age, gender, and number of children of an individual who is a target of policy recommendation from among the keywords extracted according to step (a).

In addition, in the step (a), the policy data natural language processing learning model may be trained to extract sentence components in the form of nouns from the policy announcement using morphological analysis.

In addition, the weight may be set based on a plurality of nouns extracted from the policy announcement by the policy data natural language processing learning model, and the weight of a noun with a high extraction frequency may be set to be greater than the weight of a noun with a low extraction frequency based on the frequency of the nouns extracted from the policy announcement.

In addition, the step (a) may include: (a-1) a step of setting the same vector value for a plurality of nouns determined to have the same meaning according to a preset algorithm, and setting an adjacent vector value within a preset difference from the vector value for another noun determined to have a similar meaning, thereby matching noun-vector pairs; (a-2) a step of classifying nouns extracted from the policy announcement according to a preset algorithm, whose vector values are identical or within a preset difference, into one keyword; and (a-3) a step of performing learning by setting the plurality of noun-vector pairs generated according to the step (a-1) and the keywords classified according to the step (a-2) as input values among learning values of the policy data natural language processing learning model, and setting the keywords classified according to the step (a-2) as output values among learning values of the policy data natural language processing learning model.

In addition, the weight may further include a penalty weight, and the step (b) may evaluate the relevance between the keyword extracted according to the step (a) and the category item corresponding to the keyword based on the penalty weight.

In addition, the step (c) above may be to generate the classification result table by classifying keywords related to the residence, type of business, income, age, gender, and number of children of the individual who is the target of the policy recommendation into one of the categories of the relevant organization category, policy name category, and support target category corresponding to the policy announcement according to a preset algorithm.

In addition, the step (b) may be such that, if a keyword corresponding to the above-described category item is not selected, the context of the policy announcement is re-identified using the entity recognition module and another keyword corresponding to the category item is selected.

In addition, if there are multiple keywords corresponding to the same identifier among the generated identifiers, in step (c), multiple keywords may be matched to one category item.

In addition, the step (c) may be such that the policy data natural language processing learning model matches each category item with its corresponding keyword according to a preset algorithm based on the identifier, enters the matched keyword into the corresponding category item, and generates the classification result table, and matches keywords for one item combining at least two category items, and enters the matched keyword into the corresponding item, thereby generating the classification result table.

According to one embodiment of the present invention, a tagging process performing device for recommending a policy to an individual based on natural language processed policy data includes: a memory storing a program for performing a tagging process performing a method for recommending a policy to an individual based on natural language processed policy data; and a processor executing the program, wherein the processor may be configured to perform the following: receiving keywords extracted from a policy announcement through a preset policy data natural language processing learning model, selecting keywords corresponding to preset category items from among the extracted keywords by considering weights using a named entity recognition module, generating identifiers indicating category items corresponding to the selected keywords, entering the selected keywords into category items corresponding to the keywords based on the identifiers to generate a classification result table, and generating the identifiers for keywords related to at least one of a residence, a type of business, an income, an age, a gender, and the number of children of an individual who is a target of policy recommendation from among the extracted keywords.

According to the present invention, data standardization can be provided by classifying a large amount of policy data that has undergone natural language processing into a plurality of categories.

In addition, according to the present invention, the policy data is selected as keywords based on weights corresponding to multiple category items, and through a tagging process that generates an identifier representing the category items in the policy data, the extracted policy data can be processed into a form suitable for use in industrial fields such as policy recommendation and statistical production.

Furthermore, according to the present invention, when the policy target, i.e. the user, is an individual, the policy data included in the policy announcement can be classified with a focus on individual-targeting policy data by giving weight to items targeting individuals during the tagging process.

FIG. 1 is a structural diagram of a system that automates a tagging process for recommending policies to individuals based on natural language processed policy data according to one embodiment of the present invention.
Figure 2 is a block diagram showing the internal configuration of a server according to one embodiment of the present invention.
FIG. 3a is an example diagram of a reference column among policy data according to one embodiment of the present invention.
Figure 3b is an example diagram of a target column among policy data according to one embodiment of the present invention.
FIG. 4 is an example diagram of category items according to one embodiment of the present invention.
FIG. 5 is an example diagram of a policy announcement in which an identifier is generated according to one embodiment of the present invention.
FIG. 6 is an example diagram of a category item with keywords written therein according to one embodiment of the present invention.
FIG. 7 is an example diagram of an input UI for receiving user data according to one embodiment of the present invention.
FIG. 8 is an example diagram of a group of recommended candidates displayed on a user terminal according to one embodiment of the present invention.
FIG. 9a is a flowchart illustrating a method for performing natural language processing on policy data according to an embodiment of the present invention.
FIG. 9b is a flowchart illustrating a method for performing a tagging process for recommending a policy to an individual or a company based on natural language processed policy data according to an embodiment of the present invention.
FIG. 9c is a flowchart illustrating a method for recommending policies to individuals or companies using tagged policy data according to an embodiment of the present invention.

Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily practice the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in order to clearly describe the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are assigned similar drawing reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected" but also the case where it is "electrically connected" with another element in between. Also, when a part is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise stated.

In this specification, the term 'unit' includes a unit realized by hardware, a unit realized by software, and a unit realized using both. In addition, one unit may be realized by using two or more pieces of hardware, and two or more units may be realized by one piece of hardware. Meanwhile, the '~ unit' is not limited to software or hardware, and the '~ unit' may be configured to be in an addressable storage medium and may be configured to reproduce one or more processors. Accordingly, as an example, the '~ unit' includes components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and '~sub-components' may be combined into a smaller number of components and '~sub-components' or further separated into additional components and '~sub-components'. In addition, the components and '~sub-components' may be implemented to regenerate one or more CPUs within the device or the secure multimedia card.

The "terminal" mentioned below can be implemented as a computer or portable terminal that can access a server or other terminal via a network. Here, the computer can include, for example, a notebook, desktop, laptop, VR HMD (e.g., HTC VIVE, Oculus Rift, GearVR, DayDream, PSVR, etc.) equipped with a WEB Browser. Here, the VR HMD includes all of PC-use (e.g., HTC VIVE, Oculus Rift, FOVE, Deepon, etc.), mobile-use (e.g., GearVR, DayDream, Storm Magic, Google Cardboard, etc.), and console-use (PSVR) and Stand Alone models (e.g., Deepon, PICO, etc.) that are implemented independently. Portable terminals are wireless communication devices that ensure portability and mobility, for example, and may include not only smart phones, tablet PCs, and wearable devices, but also various devices equipped with communication modules such as Bluetooth (BLE, Bluetooth Low Energy), NFC, RFID, ultrasonic, infrared, WiFi, and LiFi. In addition, "network" refers to a connection structure that enables information exchange between each node, such as terminals and servers, and includes a local area network (LAN), a wide area network (WAN), the Internet (WWW: World Wide Web), wired and wireless data communication networks, telephone networks, and wired and wireless television communication networks. Examples of wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth, infrared, ultrasonic, visible light communication (VLC), and LiFi.

The present invention relates to a method and a device for performing natural language processing on policy data, performing tagging for recommending policies to individuals or businesses based on the data for which natural language processing has been completed, and then recommending policies to individuals or businesses using the tagged policy data. The method and device are a technology for collecting predetermined information from a policy announcement and analyzing and classifying the collected information according to a preset algorithm, thereby increasing accessibility to specific policies applicable to or applicable to users.

Hereinafter, with reference to FIGS. 1 to 9c, a policy data natural language processing method according to one embodiment of the present invention, a tagging and classification method for natural language processed policy data, and a method and device for recommending a policy to an individual or a company based on the classified policy data will be sequentially described.

Referring to Figure 1, a system according to one embodiment of the present invention may be composed of a server (100), a policy agency server (200), and a user terminal (300).

Referring to FIG. 2, a server (100) according to one embodiment of the present invention may be a device including a memory storing a program (or application) that performs at least one of a method of performing natural language processing on policy data, a method of performing a tagging process for recommending a policy to an individual or a business based on data on which natural language processing has been completed, and a processor executing the above program. Here, the processor may perform various functions according to the execution of the program stored in the memory.

Next, the policy agency server (200) may be a device that is operated in a place that implements policies, such as a government agency, public agency, or private organization, and stores information on multiple policies and policy announcements currently being implemented. The policy agency server (200) may be connected to the server (100) via a wired or wireless communication network.

A user terminal (300) according to one embodiment of the present invention is capable of communicating with a server (100) by wire or wireless connection, and can be implemented in the form of a smart phone, tablet PC, PDA, and desktop.

First, the execution process of a method for performing natural language processing on policy data according to one embodiment of the present invention will be described below.

The server (100) can access the policy agency server (200) to receive a policy notice issued from the policy agency server (200), or collect the policy notice by crawling the website operated by the policy agency server (200).

The server (100) extracts multiple nouns from the collected policy announcement using a morphological analyzer.

Referring to FIG. 3a, a server (100) according to one embodiment of the present invention can divide text representing the content of a policy announcement into a reference column and a target column.

The reference column may include the policy name (service name) and policy purpose (service purpose) from among the contents included in the policy announcement.

In addition, referring to Fig. 3b, the target column may include the support target, support content, and support cost (e.g., facility cost, operating cost, and personnel cost) among the contents included in the policy announcement.

The server (100) can identify reference columns or target columns in the content of a policy announcement, extract nouns within each column, and calculate the number of times each noun appears.

At this time, the morphological analyzer that can be utilized to extract nouns in one embodiment of the present invention includes KOMORAN, KoNLPy and Khaiii morphological analyzers, and may recognize the entire text included in the policy announcement and extract multiple sentence components.

For example, when extracting nouns (sentence elements) from a policy notice regarding the support project for child care centers during the busy farming season using the KOMORAN morphological analyzer, nouns such as busy farming season, child, care, operation, and support can be extracted from the service name corresponding to the reference column among the entire text of the policy notice, and nouns such as service purpose, support content, and support target can be extracted from the target column.

Thereafter, the server (100) can generate a numerical vector for each of the multiple nouns extracted according to the process described above using an embedding technique.

At this time, according to a preset algorithm, among the extracted multiple nouns, the same vector value may be set for multiple nouns judged to have the same meaning, and for other nouns judged to have similar meanings, an adjacent vector value may be set within a preset difference from the vector value set for the corresponding nouns.

That is, the difference in vector values between multiple nouns judged to have similar meanings is set to be small, and the size of the vector values between multiple nouns with different meanings is set to be large, so that the usage and distinction between each noun can be set to be easy.

The embedding technique utilized in the present invention and the above-described preset algorithm mean the result of converting a natural language written by a person into a vector, which is a numerical form understandable by a machine, or the entire series of processes thereof, and therefore are not described in detail in this specification because they correspond to prior art.

Based on the vector values set as described above, the server (100) can generate multiple noun-vector pairs by matching the set vector values with the extracted nouns.

At this time, by generating a noun-vector pair including the sentence component located immediately before each noun and the sentence component located immediately after each noun for the nouns extracted from the target column in the policy announcement, it can be generated in a way that makes it easier to understand the context of the sentence than simply generating a pair of a single sentence component and a vector.

Next, the server (100) generates a noun-vector dictionary based on the noun-vector pairs. The noun-vector dictionary according to one embodiment of the present invention generates and stores identifiers representing vector values for nouns extracted from the contents of a plurality of policy announcements, and defines noun-vector pairs included in the noun-vector dictionary as keywords.

Keywords can be classified into multiple category items according to a preset algorithm based on data that has undergone natural language processing, which will be described later.

Next, the server (100) can set a weight for each noun-vector pair based on the distance value between each noun-vector pair included in the noun-vector dictionary and the number of times a noun is included in the policy announcement.

The weights are set based on the frequency of the extracted nouns so that the weights of nouns with a high extraction frequency are greater than the weights of nouns with a low extraction frequency. The greater the number of times a noun is included in a policy announcement, the greater the maximum weight is assigned, and the weights can be sequentially set according to preset weight intervals down to nouns with a low extraction frequency.

That is, nouns with high weights are nouns that are frequently mentioned in policy announcements, and can be judged to correspond to the core content of the policy announcements, and can be used to create a natural language processing learning model that extracts keywords from policy announcements, which will be described later.

In addition, the weights, including penalty weights, can be utilized to evaluate the relevance between keywords extracted from a policy announcement and category items corresponding to the keywords in the process of performing tagging to recommend policies to individuals based on data that has undergone natural language processing, based on the penalty weights.

Next, the server (100) can perform machine learning based on the weights and noun-vector dictionary to generate a policy data natural language processing learning model.

A policy data natural language processing learning model according to another embodiment of the present invention may include any one of a KoBERT classifier, an LSTM classifier, and an SVC method model, and may be a mixture of at least one model.

In this embodiment, each classifier included in the policy data natural language processing learning model extracts one value among multiple option values, and may include a binary classifier and a multi-classifier. Accordingly, the policy data natural language processing learning model may selectively use a binary classifier and a multi-classifier depending on the data to be extracted.

In addition, the policy data natural language processing learning model can be created by classifying nouns among a plurality of nouns extracted according to a preset algorithm, which have vector values that match or are within a preset difference, that is, nouns judged to have similar meanings, into a single keyword, and setting the noun-vector pairs and the classified keyword as input values among the learning values of the policy data natural language processing learning model, and performing learning by setting the keyword as an output value.

As described above, the policy data natural language processing learning model can be configured to automatically classify and output keywords when a new noun is input from a policy announcement based on the learning results.

In other words, after the policy data natural language processing learning model is created, when a new policy announcement is input to the server (100), the policy data natural language processing learning model can be used to extract noun-vector pairs from the new policy announcement, automatically update the noun-vector dictionary (keyword), and output the category items and the tagging results (classification results).

Referring to FIG. 9a, a method for performing natural language processing on policy data according to one embodiment of the present invention can be performed in the following order.

First, the server (100) receives a policy announcement from a policy agency server (200) and extracts multiple nouns from the policy announcement using a morphological analyzer (S101).

For the extracted nouns, a numerical vector is generated for each noun using the embedding technique, and multiple noun-vector pairs are generated (S102).

Next, the server (100) generates a noun-vector dictionary including multiple noun-vector pairs (S103).

Thereafter, based on the distance value between each noun-vector pair included in the noun-vector dictionary and the number of times it is exposed in the policy announcement, a weight is set for each noun-vector pair (S104).

The tagging process for recommending policies to individuals based on data that has undergone natural language processing through the above-described process and the method for automating the tagging process are as follows.

First, keywords extracted from the policy announcement are selected by considering weights through a named entity recognition module using a pre-established policy data natural language processing learning model, and keywords corresponding to pre-established category items are selected, and an identifier representing the category item corresponding to the selected keyword is generated.

At this time, if the target of the policy recommendation is an individual, an identifier is created for a keyword related to at least one of the individual's residence, type of business, income, age, gender, and number of children among the extracted keywords.

On the other hand, if the target of the policy recommendation is a company, an identifier is created for a keyword related to at least one of the company's sales, location, type of business, number of employees, year of establishment, and business type.

Through this, the present invention extracts sentence components of a policy announcement without regard to whether the target of the policy is an individual or a company, but in the process of tagging the extracted policy data by each category item, the tagging process can be performed by distinguishing between policies targeting individuals and companies.

Tagging in the present invention means that an identifier is generated for each extracted keyword and each category item is classified according to each identifier.

If a keyword corresponding to a preset category item is not selected, the context of the policy announcement can be re-identified using the entity name recognition module, and a process of selecting another keyword corresponding to the category item can be performed.

For example, let's say there is a policy with the service name [Support for Discovering Lifestyle Inventions] and the service purpose [Increasing the idea creation and economic activity participation of highly educated women who have taken a break from their careers to lay the foundation for new innovation and a leap forward in our economy], it is not possible to know whether this policy is only for women just from the service name, but through the contextual understanding of the policy data natural language processing learning model and the weight set for the service purpose, the corresponding keyword can be selected in the new category "Gender-Excluded" in addition to the "Gender" category.

Therefore, the present invention selects sentence components located immediately before and after an extracted noun as keywords instead of classifying category items based on an identifier generated for a single noun, thereby preventing the category item of the corresponding keyword from being classified as ‘women’ when there is a contextual policy of ‘excluding women’ because it is not read.

The contextual understanding of the policy announcement may be accomplished by re-performing the process of tagging the category items by adding the sentence elements located immediately before and immediately after the noun-vector pair corresponding to the keyword as described above, and repeating the process until the keyword corresponding to the preset category item is selected.

As illustrated in Figure 4, a category item may be composed of multiple items, such as gender, education level, job, household members, marital status, number and presence of children, type of relevant organization, type of support, application procedure, type of collection, and target characteristics. If the target of the policy recommendation is an individual, the category item may include the individual's residence, type of business, income, age, gender, and number of children. If the target of the policy recommendation is a company, the category item may additionally include the company's sales, location, type of business, number of employees, year of establishment, and business type.

Additionally, an identifier to be classified can be preset by entering it as a category item for each category item.

For example, assuming that an identifier representing a region is preset as LOC, an identifier representing the number of workers is preset as NOH, and an identifier representing a deadline is preset as DUR, a tagging process can be performed in which identifiers are generated and appended to the end of nouns extracted from a policy announcement, as shown in Fig. 5, and these identifiers can be utilized in generating a classification result table to be described later.

Next, the server (100) creates a classification result table by entering the selected keywords into category items corresponding to the keywords based on the identifier.

At this time, if the policy recommendation target is an individual, keywords related to residence, type of business, income, age, gender, and number of children are classified into one of the categories of the relevant agency category, policy name category, and support target category corresponding to the policy announcement according to a preset algorithm, and a classification result table is generated. If the policy recommendation target is a corporation, keywords related to the corporation's sales, location, type of business, number of employees, year of establishment, and business type are classified into one of the categories of the relevant agency category, policy name category, and support target category corresponding to the policy announcement according to a preset algorithm.

Referring to FIG. 6, the classification result table can match multiple keywords to one category item if there are multiple keywords corresponding to the same identifier among the generated identifiers as illustrated.

For example, if a category item is a relevant organization = Seongju-gun, Gyeongsangbuk-do, a keyword containing multiple identifiers judged to be similar to Gyeongsangbuk-do or Seongju-gun from the policy announcement can be entered as the category item and classified.

According to a further embodiment of the present invention, a classification result table can be generated by matching each category item with its corresponding keyword according to a preset algorithm based on an identifier by a policy data natural language processing learning model, which is a result of the machine learning described above, and entering the matched keyword into the corresponding category item, or a classification result table can be generated by matching keywords for one item combining at least two or more category items and merging and entering the matched keyword into the corresponding item.

Referring to FIG. 9b, the execution sequence of a method for performing a tagging process for recommending a policy to an individual or a company based on data on which natural language processing has been completed according to one embodiment of the present invention is as follows.

First, the server (100) receives keyword input extracted from a policy announcement through a preset natural language processing machine learning model (S201).

Next, using the entity name recognition module, each keyword corresponding to multiple category items is selected based on weights, and an identifier representing the category item is generated for the selected keyword (S202).

Thereafter, based on the generated identifier, multiple keywords for which the identifier was generated are entered into the corresponding category items to complete the classification result table (S203).

Based on the classification result table generated as described above, the server (100) can recommend a policy to an individual or company using the policy data for which tagging has been completed.

In addition to policy data collected from the policy agency server (200), the server (100) can collect user data and behavioral data from the access records of the user terminal (300).

Here, the policy data may include information on keywords generated through the natural language processing and tagging process described above and a classification result table.

In other words, policy data may include keywords extracted through a preset policy data natural language processing learning model from policy announcements regarding policies implemented by policy agencies, and tagging information that tags keywords and category items to classify keywords for which identifiers are generated by preset category items.

In addition, if the target of the policy recommendation is an individual, the user data may include at least one of the information on the individual's residence, type of business, income, age, gender, and number of children input from the user terminal (300).

Likewise, if the target of the policy recommendation is a company, it may include at least one of the information on the company's sales, location, type of business, number of employees, year of establishment, and business type input from the user terminal (300).

Referring to FIG. 7, a server (100) according to one embodiment of the present invention can provide an input UI to a user terminal (300) to collect necessary information depending on whether the target of the policy recommendation is an individual or a company.

Next, the behavioral data may include the connection time and number of connection sessions identified based on log data of the user terminal (300) connecting to the server (100) and the policy agency server (200).

The server (100) compares and analyzes policy data and user data to generate a recommendation candidate group including multiple policies to be recommended to the user terminal (300), and generates the recommendation candidate group so as to include policies in which the relevance of the keywords and tagging information included in the policy data and the information included in the user data matches a preset numerical value or higher.

At this time, the server (100) can select a policy corresponding to the intersection between the policy data and the user data based on the information entered by the user through the user terminal (300) to generate an initial recommendation candidate group.

Additionally, the server (100) performs a process of setting a ranking among policies included in the initial recommendation candidate group based on behavioral data.

The server (100) rearranges the rankings of the policies included in the recommended candidate group based on the sum of the access times and the number of access sessions of each policy among the policies included in the recommended candidate group, for which the access time of each user data of the behavioral data is longer than a preset time and the number of access sessions is longer than a preset number.

For example, based on how long and how many times the user terminal (300) has been connected to an Internet session providing each policy among multiple policies included in the recommendation candidate group, a policy that the user terminal (300) has connected to for a long time and frequently can be determined to be a policy that the user is interested in, and this can be used to recommend it preferentially.

If the user terminal (300) is accessing the server (100) for the first time, the server (100) can rearrange the order of policies included in the recommendation candidate group based on similar behavior data having a relevance higher than a preset value to the behavior data that occurs when the user terminal (300) first accesses the server (100) among a plurality of previously stored behavior data, and rearrangement of the order using user data is also possible.

Referring to FIG. 8, the group of recommended candidates provided to the user terminal (300) may include multiple policies as illustrated, and each policy may be provided with the type and name of the implementing agency of the policy, the policy name, and the support period displayed.

For example, if a user terminal (300) whose user data is 30-year-old, single office worker K, first accesses the server (100), the user data of a similar 30-year-old, single office worker P can be used for re-arranging the ranking, and if the behavioral data of K's user terminal (300) is determined to have accessed the Youth Tomorrow Filling Deduction Policy for a long time, the Youth Tomorrow Filling Deduction Policy can be recommended to K's user terminal (300).

Likewise, if a user terminal (300) of Group S, distribution industry, and annual sales of 10 billion KRW connects to the server (100) for the first time, similar user data of Industry Y, distribution industry and annual sales of 9 billion KRW, can be used for rearranging the ranking, and if the behavioral data of Group S user terminal (300) is determined to have connected frequently for a long time in the overseas export support business policy, the overseas export support business policy can be provided to Group S user terminal (300).

Referring to FIG. 9c, a method for recommending a policy to an individual or a company using tagged policy data according to one embodiment of the present invention can be performed in the following order.

First, the server (100) collects policy data from the policy agency server (200) and collects user data and behavioral data from the access record of the user terminal (300) (S301).

Afterwards, the policy data, for which natural language processing and tagging have been completed, is matched with user data to generate a group of recommended candidates including multiple policies (S302).

Next, the ranking of the recommended candidates is reset based on the behavioral data (S303).

Then, policies included in the sorted recommendation candidate group are recommended to the user terminal (300) in order of highest to lowest ranking (S304).

An embodiment of the present invention may also be implemented in the form of a recording medium containing computer-executable instructions, such as program modules, that are executed by a computer. The computer-readable medium may be any available medium that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may include all computer storage media. The computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer-readable instructions, data structures, program modules, or other data.

Although the methods and systems of the present invention have been described with respect to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

The above description of the present invention is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential characteristics of the present invention. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single component may be implemented in a distributed manner, and likewise, components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

100: Server 200: Policy Agency Server
300: User Terminal

Claims (10)

A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data performed by a server,
(a) A step of receiving keywords extracted from a policy announcement through a preset policy data natural language processing learning model;
(b) a step of extracting a selection keyword corresponding to a preset category item by considering the weight among the keywords extracted according to step (a) using an entity name recognition module, and generating an identifier indicating a category item corresponding to the selection keyword; and
(c) a step of generating a classification result table by entering the selection keyword in a category item corresponding to the selection keyword based on the identifier,
In the above step (b), the identifier is created for a keyword related to at least one of the residence, type of business, income, age, gender, and number of children of the individual who is the target of the policy recommendation among the keywords extracted according to the above step (a).
Step (b) above,
If the above selection keyword corresponding to the above preset category item is not extracted, the step of re-identifying the context of the policy announcement using the entity name recognition module and extracting another keyword corresponding to the above preset category item is further included.
This is performed repeatedly until a keyword corresponding to the above preset category item is extracted.
A method for performing a tagging process to recommend policies to individuals based on natural language processed policy data. In the first paragraph,
The above policy data natural language processing learning model is,
It is learned to extract sentence components in the form of nouns from the above policy announcement using morphological analysis.
A method for performing a tagging process to recommend policies to individuals based on natural language processed policy data. In the first paragraph,
The above weights are,
The above policy data is set based on multiple nouns extracted from the above policy announcement by the natural language processing learning model.
A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, wherein the weight of a noun with a high extraction frequency is set to be greater than the weight of a noun with a low extraction frequency based on the frequency of the nouns extracted from the above policy announcement. In the first paragraph,
Step (a) above,
(a-1) A step for matching noun-vector pairs by setting the same vector value for a plurality of nouns judged to have the same meaning according to a preset algorithm, and setting an adjacent vector value within a preset difference from the vector value for another noun judged to have a similar meaning to the plurality of nouns;
(a-2) a step of classifying nouns extracted from the policy announcement according to a preset algorithm into one keyword, among which nouns whose vector values match or are within a preset difference; and
(a-3) A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, comprising the steps of: setting a plurality of noun-vector pairs generated according to the step (a-1) and keywords classified according to the step (a-2) as input values among learning values of the policy data natural language processing learning model, and performing learning by setting keywords classified according to the step (a-2) as output values among learning values of the policy data natural language processing learning model. In the first paragraph,
The above weights further include penalty weights,
Step (b) above,
A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, comprising a step of evaluating the relevance of a keyword extracted according to step (a) and a category item corresponding to the keyword extracted according to step (a) based on the penalty weight. In the first paragraph,
Step (c) above,
A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, comprising the step of generating a classification result table by classifying keywords related to the residence, type of business, income, age, gender and number of children of an individual who is the target of the above policy recommendation into one of the categories of the relevant organization, the policy name and the support target corresponding to the policy announcement according to a preset algorithm. delete In the first paragraph,
If there are multiple keywords corresponding to the same identifier among the generated identifiers,
A method for performing a tagging process for recommending a policy to an individual based on natural language processed policy data, wherein, in step (c) above, multiple keywords are matched to one category item. In Article 6,
Step (c) above,
The above policy data natural language processing learning model includes a step of matching each category item with its corresponding keyword according to a preset algorithm based on the identifier, entering the matched keyword into each category item to generate the classification result table, matching a keyword for one item combining at least two or more category items, and entering the matched keyword into one item combining at least two or more category items to generate the classification result table.
A method for performing a tagging process to recommend policies to individuals based on natural language processed policy data. In a device for performing a tagging process for recommending policies to individuals based on natural language processed policy data,
A memory storing a program for performing a tagging process for recommending policies to individuals based on natural language processed policy data; and
comprising a processor for executing the above program;
The above processor,
A method of receiving keywords extracted from a policy announcement through a preset policy data natural language processing learning model, extracting selection keywords corresponding to preset category items by considering weights among the extracted keywords using a named entity recognition module, generating an identifier representing a category item corresponding to the selected keyword, entering the selected keyword in a category item corresponding to the selected keyword based on the identifier to generate a classification result table, generating the identifier for a keyword related to at least one of the residence, type of business, income, age, gender, and number of children of an individual who is a target of policy recommendation among the extracted keywords, and if the selection keyword corresponding to the preset category item is not extracted, re-identifying the context of the policy announcement using the named entity recognition module, extracting another keyword corresponding to the preset category item, and repeatedly extracting another keyword corresponding to the category item until the keyword corresponding to the preset category item is extracted.
A device that performs a tagging process to recommend policies to individuals based on natural language processed policy data.

Images