KR102657791B1 - Method for providing persona code using deep learning module and server performing the same - Google Patents

Abstract

The present invention relates to a method for generating persona codes and a server that performs the same. The persona code generation method includes receiving response data to survey data pre-stored in a database linked to the server and unstructured data collected through the user terminal from the user terminal, the answer data and the unstructured data Generating pre-processed data by performing a pre-processing operation on the pre-processed data, generating tagged data by assigning a tag corresponding to the pre-processed data, and generating a first persona code using the tag included in the tagged data. Generating a second persona code through clustering of the preprocessed data based on a category pre-specified for each survey question constituting the survey data, and generating the first persona code and the second persona code. This includes combining to generate the final persona code.

Description

Method for providing persona code using deep learning module and server performing the same}

The present invention relates to a method of providing persona code using a deep learning module and a server that performs the same. Specifically, the present invention relates to a method of generating and providing a persona code that reflects the user's tendencies in various categories in detail, and predicting and providing a persona code when user data is lacking.

The content described in this section simply provides background information for this embodiment and does not constitute prior art.

Recently, there has been an increase in the number of services that classify human tendencies into specific numbers by providing tendency analysis services using MBTI (Myers-Briggs Type Indicator). In these services, the importance of data collection to understand user tendencies is increasing significantly.

Accordingly, there was a need to collect various types of answers and create an accurate persona code that represents the user's unique tendencies. In addition, there was a need to generate highly accurate persona codes using pre-stored big data even when the user's response was insufficient.

The purpose of the present invention is to provide a persona code derived by identifying the user's characteristics based on the received user's answers and determining the user's tendency for each characteristic.

Additionally, the purpose of the present invention is to standardize the received unstructured data when receiving unstructured data including images other than text, and to generate and provide a persona code by reflecting the structured data.

Additionally, the purpose of the present invention is to derive a plurality of temporary persona codes by utilizing data on insufficient user responses, receive the user's responses to the derived persona codes, and select and provide the persona code with the highest accuracy.

In addition, the purpose of the present invention is to train a deep learning module to predict missing data using learning data including the user's answers and the final persona code, and to derive the persona code using the pre-trained deep learning module. It is provided.

The objects of the present invention are not limited to the objects mentioned above, and other objects and advantages of the present invention that are not mentioned can be understood by the following description and will be more clearly understood by the examples of the present invention. Additionally, it will be readily apparent that the objects and advantages of the present invention can be realized by the means and combinations thereof indicated in the patent claims.

A method of generating a persona code according to some embodiments of the present invention includes: response data to survey data pre-stored in a database linked to the server, and the user terminal. Receiving unstructured data collected through the user terminal, performing a preprocessing operation on the answer data and the unstructured data to generate preprocessed data, tagging each keyword included in the preprocessed data. Generating tagging data by assigning tags, Generating a first persona code using the tag included in the tagging data, Preprocessing based on a pre-specified category for each survey question constituting the survey data It includes generating a second persona code through clustering of data and combining the first persona code and the second persona code to generate a final persona code.

Additionally, the unstructured data may include at least one of images, videos, voices, and location information collected in the process of receiving user responses to the survey data at the user terminal.

In addition, the step of generating the preprocessed data includes generating extracted data consisting of preset main keywords included in the answer data through text mining on the answer data, and keywords related to objects included in the unstructured data. generating standardized data for the unstructured data by extracting the extracted data based on the reliability of each response of the survey data included in the answer data and the importance of keywords included in the structured data; and calculating a weight of the standardized data, and generating preprocessed data including the extracted data, the standardized data, and the weights for the extracted data.

In addition, the step of assigning the first persona code includes assigning first tags related to a first characteristic included in the first category of the final persona code to the preprocessed data to generate first tag assigned data; calculating a first score for the first characteristic by considering a weight of the preprocessed data for the keyword to which the first tag is assigned; a second characteristic different from the first characteristic included in the first category; Generating second tagged data by assigning related second tags to the preprocessed data, and generating a second score for the second characteristic by considering a weight for keywords to which the second tag is assigned in the preprocessed data. Calculating, assigning a first sub-persona code corresponding to the first score, assigning a second sub-persona code corresponding to the second score, the first sub-persona code and the It may include generating the first persona code by combining second sub-persona codes.

Additionally, the step of assigning the first persona code may further include correcting the first score and the second score using user information pre-entered by the user terminal.

In addition, the step of assigning the first persona code includes assigning preset tags for each topic to keywords included in the preprocessing data, and converting the tags for each topic assigned to the preprocessing data into a first characteristic or a second characteristic. Classifying, calculating a first score for the first characteristic by considering a weight of the preprocessed data for keywords tagged with the first characteristic, and assigning a tag to the second characteristic. calculating a second score for the second characteristic by considering the weight of the preprocessed data for the keyword, assigning a first sub-persona code corresponding to the first score, and assigning a first sub-persona code to the second score. It may include assigning a corresponding second sub-persona code and combining the first sub-persona code and the second sub-persona code to generate the first persona code.

In addition, the step of generating the second persona code includes classifying the preprocessed data according to each predetermined characteristic included in the second category of the final persona code, and using a different clustering model for each characteristic. Performing clustering on data with high similarity among the classified data, deriving sub-persona codes for each cluster clustered for each characteristic, and merging the plurality of sub-persona codes derived for each characteristic. Thus, the step of generating the second persona code may be included.

In addition, the step of performing the clustering includes (a) calculating a silhouette coefficient for the result of clustering data input to the clustering model for the first characteristic, and (b) calculating a clustering coefficient where the silhouette coefficient is greater than or equal to a preset reference value. selecting, (c) determining clustering of data for the first characteristic based on the selected clustering coefficient, and performing steps (a) to (c) on a second characteristic different from the first characteristic. And it may include repeating the same steps for the third characteristic.

In addition, determining whether missing data exists in the data received from the user terminal, and if the missing data exists, inputting preprocessing data for the received data into a pre-trained deep learning module, and receiving at least one temporary persona code from the deep learning module as an output, deriving new survey data related to the received at least one temporary persona code, and providing the survey data to the user terminal; , receiving response data for the survey data from the user terminal, deriving reliability of the temporary persona code based on the response data, and determining a final persona code based on the reliability. You can.

In addition, storing the pre-processing data for the received data that does not include the missing data and the final persona code derived based on the pre-processing data in the database, and storing the pre-processing data stored in the database It may further include supervised learning of the deep learning module by applying it to the input node of the deep learning module and applying the persona code for the preprocessed data to the output node of the deep learning module.

Meanwhile, a server according to some embodiments of the present invention includes a processor, a memory for loading a computer program executed by the processor, and an interface for exchanging data generated during the execution of the computer program with a user terminal, The computer program includes receiving response data for survey data pre-stored in a database linked to the server and unstructured data collected through the user terminal from the user terminal, preprocessing the response data and the unstructured data. Generating preprocessed data by performing an operation, Generating tagging data by assigning a tag to each keyword included in the preprocessing data, Creating a first persona using the tag included in the tagging data Generating a code, generating a second persona code through clustering of the preprocessed data based on categories pre-specified for each survey question constituting the survey data, and the first persona code and the second persona code It includes the step of combining to generate the final persona code.

A method of providing a persona code according to some embodiments of the present invention generates preprocessed data using response data to a received user's questionnaire, and generates preprocessed data of the user corresponding to a plurality of characteristics divided into predetermined categories through the preprocessed data. By deriving and merging each persona code, it is possible to provide a persona code that reflects the user's hyper-personalized data.

In addition, the present invention provides a method of preprocessing unstructured data in a different form from answer data and generating a persona code using the generated preprocessed data, so that the user's preference can be more precisely reflected in the persona code.

In addition, the present invention generates a plurality of temporary persona codes using a pre-trained deep learning model when the user's response data is insufficient (i.e., when the received user data is missing data) and generates additional user responses. By providing a persona code derived by verifying a temporary persona code created using data, the prediction accuracy of the persona code can be improved.

In addition, the present invention trains a deep learning model through supervised learning using highly accurate learning data consisting of persona codes derived based on user questionnaire response data and unstructured data, and uses the learned deep learning model to You can use this to derive and use a highly accurate persona code for missing data.

In addition to the above-described content, specific effects of the present invention are described below while explaining specific details for carrying out the invention.

1 is a conceptual diagram illustrating a system that performs a method for providing a persona code according to some embodiments of the present invention.
FIG. 2 is a block diagram illustrating the configuration of the server of FIG. 1 by way of example.
Figure 3 is a flowchart illustrating a method for providing a persona code according to some embodiments of the present invention.
Figure 4 is a block diagram illustrating a method for providing a persona code according to some embodiments of the present invention.
FIG. 5 is a diagram for explaining the first and second persona codes that make up the final persona code of FIG. 3.
FIG. 6 is a diagram for explaining the categories and contents indicated by the first and second personas of the final persona code of FIG. 5.
FIG. 7 is a diagram illustrating an example of survey data included in the category of FIG. 6 and response data corresponding to the survey data.
Figure 8 is a flowchart showing the process of generating the preprocessing data of Figure 3.
FIG. 9 is a block diagram showing the specific configuration of a preprocessing module that generates the preprocessing data of FIG. 8.
FIG. 10 is a diagram showing an example of unstructured data of FIG. 8.
FIG. 11 is a diagram illustrating an example of structured data that has undergone a preprocessing operation on the unstructured data of FIG. 10.
FIG. 12 is a flow chart illustrating an embodiment of the process of generating the first persona code of FIG. 3.
FIG. 13 is a block diagram specifying the configuration of the tag module and first persona code generation module of FIG. 12.
FIG. 14 is a flowchart showing another embodiment specifying the process of generating the first persona code of FIG. 3.
FIG. 15 is a block diagram specifying the configuration of the tag module and first persona code generation module of FIG. 14.
FIG. 16 is a diagram showing an example of assigning a tag in FIG. 15.
FIG. 17 is a flowchart detailing the process of generating the second persona code of FIG. 3.
FIG. 18 is a block diagram showing the configuration of the second persona code generation module of FIG. 17.
Figure 19 is a flowchart detailing the process of performing clustering in Figure 17.
FIG. 20 is a diagram for explaining the clustering process of FIG. 19.
Figure 21 is a flowchart showing a method for generating a persona code when missing data is included according to some embodiments of the present invention.
Figure 22 is a block diagram illustrating a method for generating a persona code when missing data is included according to some embodiments of the present invention.
FIG. 23 is a diagram for explaining the structure of the persona code generation deep learning module of FIG. 22.
Figure 24 is a flowchart for explaining the learning method of the persona code deep learning module of Figure 22.
Figure 25 is a block diagram for explaining the learning method of the persona code deep learning module of Figure 22.
Figure 26 is a diagram for explaining the hardware configuration of a server that performs a persona code generation method according to some embodiments of the present invention.

Terms or words used in this specification and patent claims should not be construed as limited to their general or dictionary meaning. According to the principle that the inventor can define the term or word concept in order to explain his or her invention in the best way, it should be interpreted with a meaning and concept consistent with the technical idea of the present invention. In addition, the embodiments described in this specification and the configurations shown in the drawings are only one embodiment of the present invention and do not completely represent the technical idea of the present invention, so they cannot be replaced at the time of filing the present application. It should be understood that there may be various equivalents, variations, and applicable examples.

Terms such as first, second, A, and B used in the present specification and claims may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be named a second component without departing from the scope of the present invention, and similarly, the second component may also be named a first component. The term 'and/or' includes any of a plurality of related stated items or a combination of a plurality of related stated items.

The terms used in the specification and claims are merely used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "include" or "have" should be understood as not precluding the existence or addition possibility of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification. .

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by a person of ordinary skill in the technical field to which the present invention pertains.

Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present application. No.

Additionally, each configuration, process, process, or method included in each embodiment of the present invention may be shared within the scope of not being technically contradictory to each other.

Hereinafter, a method for providing a persona code and a system for performing the method according to some embodiments of the present invention will be described in detail with reference to FIGS. 1 to 26.

1 is a conceptual diagram illustrating a system that performs a method for providing a persona code according to some embodiments of the present invention.

Referring to FIG. 1, a system according to some embodiments of the present invention includes a platform server 100 (hereinafter referred to as server) and a user terminal 200. Here, the server 100 may operate in conjunction with the user terminal 200 through the communication network 300.

The server 100 becomes the entity that performs the persona code provision method according to some embodiments of the present invention. However, the present invention is not limited to this, and of course, in some embodiments of the present invention, the user terminal 200 can perform some of the persona code providing method of the present invention.

The user terminal 200 refers to a terminal that receives response data and unstructured data from a user for a survey used in the persona code provision method provided by the server 100. At this time, the user terminal 200 may include an imaging device (eg, camera) that photographs the user's appearance or the surrounding environment. Additionally, the user terminal 200 may include a location tracking device (eg, GPS device) that determines the user's location information.

Specifically, in one embodiment of the present invention, the user terminal 200 may receive answer data and unstructured data from the user through an input interface. However, the above description is only a few examples of transmitting answer data and unstructured data from the user terminal 200, and the present invention is not limited thereto.

The user terminal 200 may transmit the user's answer data and unstructured data to the server 100 for pre-stored survey data. Additionally, according to an embodiment of the present invention, the user terminal 200 may transmit only one of answer data and unstructured data to the server 100.

At this time, the user terminal 200 may refer to various electronic devices such as a personal computer (PC), a laptop, a tablet, a mobile phone, a smartphone, or a wearable device (eg, a watch-type terminal). In addition, the user terminal 200 includes an input unit (not shown) that receives the user's input, a display unit (not shown) that displays visual information, a communication unit (not shown) that transmits and receives signals to and from the outside, and a data processing unit that processes the user terminal. (200) It may include a control unit (not shown) that controls each internal unit and controls data transmission/reception between units.

The server 100 may provide a persona code generation service to users who use the user terminal 200 through an online platform. These online services provided by the server 100 may be provided through an application pre-installed on the user terminal 200.

Below, we will take a closer look at the configuration of the server 100 that performs the persona code provision method.

FIG. 2 is a block diagram illustrating the configuration of the server of FIG. 1 by way of example.

Referring to FIG. 2, the server 100 of the present invention includes an interface 110, a database 120, a processor 130, and a memory 140. At this time, the memory 140 includes a preprocessing module (PM), a tagging module (TM), a code generating module (CGM), and a verification module (Verifying Module). VM) may be loaded and driven or executed by the processor 130. Each module may be stored and used in the form of a computer program in the database 120 or storage (not shown) included in the server 100.

At this time, the code generation module (CGM) includes the first persona code generating module (Persona Code Generating Module 1; hereinafter, PCGM1), the second persona code generating module (Persona Code Generating Module 2; hereinafter, PCGM2), and persona code generation deep learning. It may include a module (Persona Code Deep Learning Module; hereinafter referred to as PCDM).

Additionally, a training module (TRM) may be additionally loaded into the memory 140 and driven or executed by the processor 130 according to an embodiment of the present invention. A detailed description of each embodiment will be provided below.

First, the interface 110 performs a function of exchanging data with the user terminal 200. The interface 110 may provide survey data pre-stored in the user terminal 200 and receive the user's response data and unstructured data for the provided survey data. At this time, the interface 110 may transmit the received answer data and unstructured data to other components within the server 100. The interface 110 may include various communication modules and may exchange data with the user terminal 200 through the communication network 300.

The database 120 performs the function of storing and managing data received through the interface 110. The database 120 may be loaded into the memory 140 and store a plurality of available modules in program form.

Additionally, the database 120 may store and manage not only training data for training a persona code generation deep learning module (PCDM), but also weights of a backbone network composed of a plurality of layers. However, this is only one example, and the present invention is not limited thereto.

The processor 130 may execute software to control at least one other component (eg, hardware or software component) of the server 100 and may perform various data processing and calculations. For example, the processor 130 loads information, commands, or data received from another component (e.g., the interface 110) into the memory 140 and performs an operation using the loaded information, commands, or data. And the resulting data can be stored in memory 140 or storage (not shown).

Additionally, the processor 130 may derive keywords through web crawling. The processor 130 may store a table matching the derived keywords and the user's personality in the database 120.

The memory 140 may store and load various data used in at least one component (eg, processor 130) of the server 100. For example, data may include input data or output data for software and instructions related thereto.

Accordingly, the processor 130 may load and use modules or instructions related to various operations of the method for providing persona codes according to some embodiments of the present invention onto the memory 140.

Specifically, the processor 130 may preprocess the answer data or unstructured data received through the interface 110 using the preprocessing module (PM) loaded on the memory 140 and transmit it to the tag module (TM). there is.

At this time, the tag module (TM) may generate tagged data by assigning a tag corresponding to the preprocessed data.

Figure 3 is a flowchart illustrating a method for providing a persona code according to some embodiments of the present invention. Figure 4 is a block diagram illustrating a method for providing a persona code according to some embodiments of the present invention. Hereinafter, the method for providing a persona code according to some embodiments of the present invention will be described by taking as an example that the performer is the server 100 or the above-described processor 130 included in the server 100.

First, referring to FIGS. 3 and 4, in the method of providing a persona code according to some embodiments of the present invention, the server 100 receives the user's response data and unstructured data for pre-stored survey data from the user terminal 200. Can receive (S100).

At this time, the answer data transmitted from the user terminal 200 may include multiple-choice, short-answer, and descriptive answers. Additionally, unstructured data may be collected in the process of receiving user responses to survey data at the user terminal 200. Specifically, unstructured data may include at least one of images, video, audio, and location information.

Specifically, the processor 130 of the server 100 may extract pre-stored survey data from the database 120. Subsequently, the processor 130 may provide survey data to the user terminal 200. At this time, the survey data may include questions that can determine the user's tendency.

The processor 130 may receive user response data and unstructured data regarding survey data. At this time, both answer data and unstructured data can be received. Specifically, the answer data may be multiple-choice or short-answer answers. Additionally, unstructured data may be data that has a form other than text. Survey data, response data, and unstructured data will be explained in detail later.

Next, the server 100 generates preprocessing data by performing a preprocessing operation on the received answer data and unstructured data (S200). Specifically, the preprocessing module (PM) of the server 100 may generate preprocessing data using at least one of answer data and unstructured data. Preprocessed data may be data in which keywords are extracted from response data and weighted. Additionally, the preprocessed data may be data in which unstructured data is standardized and weighted.

Next, the server 100 may assign a tag corresponding to the preprocessed data (S300). Specifically, the tag module (TM) of the server 100 can automatically assign tags to preprocessed data representing the user's tendencies. That is, when the preprocessed data is text, the tag module (TM) can distinguish the preprocessed data including the main text using a document analysis technique. For example, the tag module (TM) can use TF-IDF (Term Frequency-Inverse Document Frequency) to analyze the frequency of the text and the scarcity of the text within the document set and determine it to be a main text. Subsequently, the tag module (TM) can generate tagged data by assigning tags to key texts.

Next, the server 100 may generate a first persona code using the assigned tag (S400). The first persona code can be created using a tag included in tag assignment data.

Subsequently, the server 100 may generate a second persona code through clustering of the preprocessed data based on the category specified in the survey data (S500). Here, the category of each survey question may be pre-designated in the database 120. At this time, the first persona code and the second persona code may be generated in parallel. That is, the code generation module (CGM) of the server 100 may first generate either the first persona code or the second persona code. Additionally, the code generation module (CGM) can simultaneously generate the first persona code and the second persona code.

Next, the server 100 may generate a final persona code by combining the first persona code and the second persona code (S600). The final persona code may contain letters or numbers. At this time, each letter or number may represent the user's personality.

FIG. 5 is a diagram for explaining the first and second persona codes that make up the final persona code of FIG. 3. FIG. 6 is a diagram for explaining the categories and contents indicated by the first and second personas of the final persona code of FIG. 5.

Referring to FIG. 5, the final persona code may include at least one persona code. At this time, at least one persona code may include first and second persona codes. Specifically, the first persona code may represent the user's tendency related to one category. Categories may include Original Nature Factor, Environmental Factor, and Moment Factor. Additionally, one category may include multiple characteristics.

Specifically, referring to FIG. 6, the nature element may refer to an element that can specify a person's personality based on integrated psychological theory. In other words, the nature factor can represent the user's personality that can be analyzed by disposition theory. Nature elements may represent, for example, MBTI (Myers-Briggs Type Indicator), DISC, and Platinum Rule.

Specifically, nature factors can utilize multiple behavioral traits according to integrative psychological theory. In other words, the nature factor can refer to a long-term personality that represents a person's nature that does not change easily.

First, nature factors can be divided into extroversion versus introversion, depending on how you interact with people and where you get your energy from. Next, the nature factor can be divided into sensing versus intuition, depending on how information is collected and understood, and thinking versus feeling, depending on how logic and emotions are used in the decision-making process. Additionally, the nature factor can be divided into judgment versus perception depending on one's attitude toward the external world and lifestyle. However, this embodiment is not limited to this.

The first persona code and the second persona code may include a plurality of sub-persona codes. Specifically, the first persona code may include first sub-persona codes to fourth persona codes. At this time, the first sub-persona code may indicate whether the user's personality is extroverted or introverted. For example, if the first sub-persona code is 0, it may mean that the user is an extrovert, and if the first sub-persona code is 1, it may mean that the user is an introvert.

Meanwhile, the second sub-persona code may indicate whether the way the user collects and understands information relies on senses or intuition. For example, if the second sub-persona code is 0, it may mean that the user follows their senses, and if the second sub-persona code is 1, it may mean that the user follows their intuition.

Additionally, the third sub-persona code can indicate whether logic or emotion is used in the decision-making process. For example, if the third sub-persona code is 0, it may mean that the user follows reason, and if the third sub-persona code is 1, it may mean that the user follows emotions. Subsequently, the fourth sub-persona code can indicate whether the attitude toward the external world relies on judgment or perception. For example, if the fourth sub-persona code is 0, it may mean that the user's lifestyle depends on judgment, and if the fourth sub-persona code is 1, it may mean that the user's lifestyle depends on recognition.

Referring again to FIG. 5, the second persona code may include environmental elements and momentary elements. Environmental factors can specify psychology such as values and desires that change depending on environmental changes. In addition, momentary elements can specify momentary emotions, tastes, psychological tendencies, etc. according to specific situations and trends. In other words, environmental factors can refer to the user's personality collected from a mid-term perspective, and momentary factors can refer to the user's personality collected from a short-term perspective.

Additionally, the second persona code may include a plurality of sub-persona codes. Each sub-persona code may correspond to each characteristic.

Specifically, referring to FIG. 6, an environmental element may include a plurality of characteristics within a category. Specifically, characteristics included in environmental factors may include types of values and desires for home, work, and social. For example, environmental factors may include marital status and child status in relation to home, and may include name and occupation in relation to work. Additionally, environmental factors may include the presence or absence of friends of the opposite sex in relation to social networking.

Meanwhile, a moment element may include multiple characteristics within a category. Specifically, characteristics included in moment elements may include types of dispositions for emotions, situations, and tastes. For example, moment elements may include joy, sadness, and anger in relation to emotions, and weather and anniversaries in relation to situations. Additionally, in-the-moment elements may include tastes and current trends in relation to tastes.

FIG. 7 is a diagram illustrating an example of survey data included in the category of FIG. 6 and response data corresponding to the survey data.

Referring to FIGS. 2 and 7 , characteristics corresponding to categories and survey data corresponding to each characteristic may be stored in the database 120 of the server 100. Specifically, if the categories are divided into nature, environment, and moment factors, characteristics within the environment factor may include home, children, work, and standard of living. Survey data may include questions about marital status, when one wishes to marry, and the reasons for wishing to marry, in response to family characteristics. The answer data may be a multiple-choice answer for marital status, such as single or married. As another example, the response data may be a short answer that describes the age or desired year regarding the desired time to get married. As another example, the response data may be a descriptive answer that specifically describes the user's view on marriage regarding the reasons for wanting to get married.

Figure 8 is a flowchart showing the process of generating the preprocessing data of Figure 3. FIG. 9 is a block diagram showing the specific configuration of a preprocessing module that generates the preprocessing data of FIG. 8. FIG. 10 is a diagram showing an example of unstructured data of FIG. 8. FIG. 11 is a diagram illustrating an example of structured data that has undergone a preprocessing operation on the unstructured data of FIG. 10.

Referring to Figures 8 and 9, the preprocessing module (PM) used in some embodiments of the present invention may include a text mining model (TMM), a normalization model (NM), and a weight calculation model (WCM). The text mining module (TMM) of the preprocessing module (PM) can generate extracted data consisting of only the main keywords included in the answer data through text mining on the answer data (S210).

Answer data may include multiple-choice or short-answer text. Additionally, the answer data may include sentence-type text. At this time, the database 120 of the server 100 may store preset main keywords. The database 120 may include key keywords that indicate user tendencies. The text mining module (TMM) can perform text mining to derive key keywords by crawling response data. Specifically, the text mining module (TMM) can recognize contained words using Python's pandas. Subsequently, the text mining module (TMM) can recognize key keywords included in the answer data by comparing the crawled words with the pre-stored table. Text mining, for example, may utilize a morphological analysis algorithm to compare crawled words with a table previously stored in the database 120. In other words, the text mining module (TMM) can generate extracted data consisting of only key keywords included in both pre-stored tables and preprocessed data.

At this time, the pre-stored table may include new words and words with ambiguous meanings. That is, the database 120 can separately classify and store new words and words with ambiguous meanings.

Next, the structured model (NM) included in the preprocessing module (PM) can generate structured data based on unstructured data (S220). Specifically, the server 100 may extract keywords related to objects included in the unstructured data and generate structured data for the unstructured data.

Referring to FIG. 9, unstructured data may include image data, video data, voice data, and location information data. At this time, unstructured data may include data about the environment surrounding the user. However, these are only a few examples of unstructured data, and the present invention is not limited to these.

Specifically, referring to FIG. 10 , image data may include images that serve as a basis for determining a user's tendency. Image data may include, for example, images of food or drinks consumed by the user. Additionally, image data may include clothing, shoes, etc. worn by the user. In addition, image data may include images of spaces that reflect the user's tastes. However, this embodiment is not limited to this.

Referring again to FIG. 9, voice data may include the user's preferred music or ambient noise. Likewise, image data may include the user's surrounding environment.

Meanwhile, unstructured data may include data representing user reactions. Specifically, image data or video data may include emotions such as joy, sadness, surprise, anger, fear, and embarrassment in the user's reaction. That is, image data or video data may include non-verbal expressions (eg, posture, facial expression, gaze).

Additionally, voice data may include the user's speaking habits or the level of language used. Specifically, speech data may include semi-linguistic expressions (e.g., pronunciation, articulation, pitch, tone, speed). Additionally, location data may include the user's preferred location or the user's main consumption location. In other words, unstructured data may include data in a format other than text, related to the user's tendencies. However, this embodiment is not limited to this.

A formal model (NM) can generate structured data from unstructured data. Specifically, a structured model (NM) can represent unstructured data as text. In other words, structured data is data that expresses the contents of unstructured data in text. Referring to FIG. 11, when the unstructured data is the user's image data, the normalization module (NM) can derive the numerical value of each part of the face from the image data as standardized data. Additionally, standardized data may include emotions derived from a combination of values for each part of the face.

Referring again to FIG. 8, the weight calculation model (WCM) included in the preprocessing module (PM) calculates the weights of the extracted data and standardized data based on the reliability of each response included in the answer data and the importance of keywords included in the standardized data. can be calculated (S230). In other words, the weight of the extracted data and the standardized data can be calculated based on the reliability of each response in the survey data included in the answer data and the importance of the keywords included in the standardized data.

Specifically, the reliability of each response included in the response data may be determined according to the value of the data. In other words, reliability can be determined through continuous data collection and inspection. For example, if there are two or more survey data on the same characteristics in the same category, the reliability may be determined to be low if contradictory answers are given to each survey. On the other hand, if you provide consistent answers to each questionnaire, reliability can be determined to be high.

As another example, the reliability of each response may be determined based on the number of updates and update timing of the response data. If the answer data is updated frequently and recently, reliability can be determined to be high. On the other hand, if the answer data is rarely updated and has been updated in the past, the reliability may be determined to be low.

Meanwhile, importance can be derived based on keywords derived from unstructured data. The importance of words included in structured data can be determined depending on the type of unstructured data. Specifically, when the unstructured data is image data, importance may be determined according to the proportion of images corresponding to keywords derived from the image data. For example, if images corresponding to keywords occupy most of the image data, their importance may be determined to be high. On the other hand, if the image corresponding to the keyword occupies only a portion of the image data, its importance may be determined to be low.

As another example, importance can be determined based on the number of updates and timing of updates of unstructured data. Specifically, if unstructured data is frequently updated and was recently updated, reliability may be determined to be high. On the other hand, if unstructured data is updated infrequently and has been updated in the past, reliability may be determined to be low.

Subsequently, the preprocessing module (PM) may generate preprocessing data including the calculated weights (S240). In other words, it may generate preprocessing data including extracted data and standardized data and weights therefor. Specifically, in the case of response data, preprocessing data may include extracted data and weights based on the reliability of each response. Additionally, in the case of unstructured data, preprocessed data may include weights based on the importance of structured data and keywords.

In summary, the preprocessed data may include extracted data derived based on answer data and a first weight thereof, and structured data derived based on unstructured data and a second weight thereof. At this time, of course, the first weight and the second weight can be adjusted based on a predetermined ratio for answer data and unstructured data. However, this is only an example of the present invention, and the weights included in the preprocessing data can be calculated and used in various ways.

FIG. 12 is a flow chart illustrating an embodiment of the process of generating the first persona code of FIG. 3. FIG. 13 is a block diagram specifying the configuration of the tag module and first persona code generation module of FIG. 12.

Referring to FIGS. 12 and 13 , the first persona creation module (PGCM1) may include first to fourth score calculation models (SCM1 to 4) and a score correction model (SRM). The tag module TM may generate first tag assigned data by assigning first tags related to the first characteristic included in the first category to the preprocessed data (S410). Categories for user tendencies may include first to fourth categories. One category may include at least one characteristic. That is, the first category may include the first to fourth characteristics.

The tag module (TM) can assign tags according to pre-stored categories and characteristics. Tags can be automatically assigned to key keywords included in preprocessed data. Specifically, the tag module (TM) may assign first tags related to the first characteristic to the preprocessed data. At this time, the tag module (TM) may generate first tagged data by assigning a first tag to the preprocessed data. Subsequently, the first tag assignment data may be transmitted to the first persona creation module (PCGM1).

At this time, the first persona generation module (PCGM1) may include first to fourth score calculation models (SCM1 to 4) and a score correction model (SRM). That is, the first score calculation model (SCM1) can receive first tag assignment data.

Subsequently, the first score calculation model (SCM1) may calculate the first score for the first characteristic by considering the weight of the tagged preprocessed data (S420). At this time, the first score may be a value obtained by multiplying the first tagged data by a weight. Additionally, when the weight is an integer, it is considered that there is as much first tagged data as the weight, and the first score can be expressed as the number of first tagged data. However, the embodiments of the present invention are not limited thereto.

Subsequently, the second score calculation model (SCM2) may generate second tag assigned data by assigning second tags related to the second characteristic different from the first characteristic included in the first category to the preprocessed data (S430). Specifically, the tag module (TM) may assign second tags related to the second characteristic to the preprocessed data. At this time, the tag module (TM) may generate second tagged data by assigning a second tag to the preprocessed data. Subsequently, the second tag assignment data may be transmitted to the first persona creation module (PCGM1).

That is, the second score calculation model (SCM2) can receive the second tag assignment data. Likewise, the third score calculation model (SCM3) may receive third tag assignment data, and the fourth score calculation model (SCM4) may receive fourth tag assignment data.

Next, a second score for the second characteristic may be calculated by considering the weight of the preprocessed data of the keyword to which the second tag is assigned (S440).

At this time, the second score may be a value obtained by multiplying the second tag assigned data by a weight. Additionally, when the weight is an integer, it is considered that there is as much second tagged data as the weight, and the second score can be expressed as the number of second tagged data. Likewise, the third and fourth scores can be calculated based on the third and fourth tag assignment data, respectively.

Next, the score correction model (SRM) may correct the first score and the second score calculated based on user information pre-entered by the user (S450). Specifically, user information may refer to the user's personal information entered when registering as a member. For example, user information may include the user's gender, name, age, or address. At this time, the score correction model (SRM) can correct the first and second scores using user information. However, of course, in some embodiments of the present invention, step S250 may be omitted.

Next, the first sub-persona code and the second sub-persona code corresponding to the first score and the second score may be assigned (S460). Specifically, the first sub-persona code may be a one-digit number or letter corresponding to the first score. Additionally, the second sub-persona code may be a one-digit number or letter corresponding to the second score.

Meanwhile, the first persona code generation module (PCGM1) may generate a third sub-persona code and a fourth sub-persona code corresponding to the third score and the fourth score, respectively. However, embodiments of the present invention are not limited thereto.

Next, the first and second sub-persona codes can be combined to generate the first persona code (S470). Additionally, the first persona code can be generated by further combining the third and fourth sub-persona codes with the first and second sub-persona codes. At this time, the first persona code may include first to fourth sub-persona codes.

FIG. 14 is a flowchart showing another embodiment specifying the process of generating the first persona code of FIG. 3. FIG. 15 is a block diagram specifying the configuration of the tag module and first persona code generation module of FIG. 14. FIG. 16 is a diagram showing an example of assigning a tag in FIG. 15. In the following, content that overlaps with the above-mentioned content will be omitted and the differences will be mainly explained.

Referring to FIGS. 14 to 16, tags by subject can be assigned to preprocessed data (S415). The server 100 may assign preset tags for each topic to keywords included in the preprocessed data. At this time, the preset topic may be a classification standard that can group keywords. In other words, the preset subject may include first and second characteristics. Additionally, the database 120 of the server 100 may store topics that may be associated with keywords in advance.

For example, if the preprocessed data includes 'lively', 'sociable', 'friendly', and 'emotional' as keywords, the topic may be personality. As another example, if the preprocessed data includes 'night owl', 'lots of gatherings', 'living with family', 'pilates', and 'enjoying domestic travel' as keywords, the topic may be lifestyle.

Next, thematic tags assigned to the preprocessed data can be classified into first characteristics or second characteristics (S425). Tag classification and scoring model (TDSM) can classify tagged data by characteristics. Specifically, the tag classification and scoring model (TDSM) can classify the contents of preprocessed data corresponding to tagged data for a topic by characteristics. At this time, the first characteristic and the second characteristic may exhibit different characteristics. For example, the first characteristic may mean personality, and the second characteristic may mean lifestyle.

In some embodiments, multiple subjects may represent a characteristic. For example, the topic is personality and lifestyle, and the two topics can represent one characteristic, that is, the user's tendency to be extroverted or introverted.

Next, the first score for the first characteristic can be calculated by considering the weight of the preprocessed data tagged for the first characteristic (S435). The tag classification and scoring model (TDSM) can calculate the first score by considering the weight of the preprocessed data. At this time, the first score may be a value obtained by multiplying the number of tags classified as majority by a weight.

Specifically, when the first characteristic is personality, keyword data indicating an extroverted tendency and keywords indicating an introverted tendency may exist. For example, 'active', 'sociable', and 'friendly' can be classified as keywords indicating an extroverted tendency, while 'emotional' and 'subjective' can be classified as keywords indicating an introverted tendency. there is. At this time, since there are three tags corresponding to keywords indicating extroverted tendencies and two tags corresponding to keywords indicating introverted tendencies, the tags indicating extroverted tendencies may account for the majority. At this time, the first score may be 3* (weight). However, the embodiments of the present invention are not limited thereto.

Next, a second score for the second characteristic may be calculated by considering the weight of the preprocessed data tagged for the second characteristic (S445). Likewise, the tag classification and scoring model (TDSM) can calculate the second score by considering the weight of the preprocessed data. At this time, the second score may be a value obtained by multiplying the number of tags classified as majority by a weight.

Specifically, when the second characteristic is a lifestyle, there may be a keyword indicating a social tendency and a keyword indicating an independent tendency. For example, ‘lots of gatherings’, ‘living with family’, and ‘enjoying domestic travel’ can be classified as keywords that indicate social tendencies. On the other hand, 'night owl' and 'Pilates' can be classified as keywords that indicate an independent tendency. However, embodiments of the present invention are not limited thereto.

At this time, since there are four tags corresponding to keywords indicating social tendencies and two tags corresponding to keywords indicating introverted tendencies, the tags indicating social tendencies may account for the majority. At this time, the second score may be 4* (weight).

Subsequently, the server 100 may perform steps S450 to S470 described above.

Hereinafter, a method for generating a second persona code in some embodiments of the present invention will be described in detail.

FIG. 17 is a flowchart detailing the process of generating the second persona code of FIG. 3. FIG. 18 is a block diagram showing the configuration of the second persona code generation module of FIG. 17.

Referring to FIGS. 17 and 18 , the second persona code generation module (PCGM) may include a classification model (DM) and at least one sub-persona code generation module. Specifically, the second persona code generation module (PCGM) may include a first sub-persona code generation module (SPCM1) to an eighth sub-persona code generation module (SPCM8).

The server 100 may classify the preprocessed data according to each predetermined characteristic included in the second category (S510). That is, the server 100 may classify the preprocessed data according to each predetermined characteristic included in the second category of the final persona code.

At this time, the database 120 of the server 100 may include at least one characteristic in the second category. The second category may include, for example, the fifth to twelfth characteristics. Here, the fifth to twelfth characteristics may be characteristics related to environmental factors and momentary factors described above with reference to FIG. 6, but the present invention is not limited thereto.

Subsequently, the classification model (DM) may classify the received preprocessed data to correspond to the fifth to twelfth characteristics, respectively.

Next, the server 100 may perform clustering on data with high similarity among the classified data using different clustering models for each characteristic (S520).

Subsequently, the server 100 may derive sub-persona codes for clusters clustered for each characteristic (S530). Specifically, the sub-persona code generation module can derive sub-persona codes for each characteristic from the cluster containing the most data. For example, the first sub-persona code generation module (SPCM1) may derive the fifth sub-persona code corresponding to the fifth characteristic from the cluster containing the largest amount of data. In the same way, the 6th to 12th sub-persona codes can be derived through the 2nd to 8th sub-persona code generation modules (SPCM2 to 8), respectively.

Next, the server 100 may generate a second persona code by merging the sub-persona codes derived for each characteristic (S540). That is, the second persona code generation module (PCGM2) may generate the second persona code by merging the fifth to twelfth sub-persona codes corresponding to the fifth to twelfth characteristics. At this time, the second sub-persona code generation module (PCGM2) may sequentially merge the 5th to 12th sub-persona codes. However, embodiments of the present invention are not limited thereto.

Figure 19 is a flowchart detailing the process of performing clustering in Figure 17. FIG. 20 is a diagram for explaining the clustering process of FIG. 19.

18 to 20, each sub-persona code generation module may include a clustering coefficient calculation model and a clustering model. Hereinafter, the first sub-persona code generation module (SPCM1) will be described as an example.

In some embodiments, the first sub-persona code generation module (SPCM1) may include a first clustering coefficient calculation model (CVM1) and a first clustering model (CM1). The first clustering coefficient calculation model (CVM1) may receive preprocessing data for the fifth characteristic.

Subsequently, the server 100 may calculate a silhouette coefficient for the data clustering result input into the clustering model for a specific characteristic (S521). In other words, the silhouette coefficient for the data clustering result input into the clustering model for the fifth characteristic can be calculated (step (a)). Specifically, the first clustering coefficient calculation model (CVM1) may calculate a silhouette coefficient corresponding to the clustering model for the fifth characteristic. The silhouette coefficient is an indicator that evaluates the clustering results and can have values between -1 and 1. At this time, the closer the silhouette coefficient is to 1, the further apart each cluster is, meaning that clustering is done well. Additionally, the closer the silhouette coefficient is to 0, the closer each cluster is to each other. On the other hand, if the silhouette coefficient has a negative value, it means that the clustering was done incorrectly.

Subsequently, the server 100 may select a clustering coefficient (k) whose silhouette coefficient is greater than or equal to a preset reference value (S523, step (b)). At this time, the clustering coefficient (k) may mean the number of clusters. The server 100 can select the clustering coefficient (k) using a graph. Specifically, the server 100 may utilize an algorithm to use the number of clusters in a section where the sum of distances between clusters rapidly decreases as the clustering coefficient (k). At this time, the clustering coefficient (k) can be selected when the silhouette coefficient is 0.7 or more. The first clustering coefficient calculation model (CVM1) may transmit the clustering coefficient (k1) to the first clustering model (CM1). However, embodiments of the present invention are not limited thereto.

Subsequently, the server 100 may confirm clustering of data for specific characteristics based on the selected clustering coefficient (S525). Specifically, clustering of data for the fifth characteristic can be confirmed based on the selected clustering coefficient (step (c)). For example, the first clustering model (CM1) may generate a clustering graph for the fifth characteristic according to the clustering coefficient (k1).

Subsequently, the server 100 may repeatedly perform steps (a) to (c) for characteristics different from the fifth characteristic. That is, the server 100 may generate each sub-persona code by performing steps (a) to (c) for the sixth to twelfth characteristics that are different from the fifth characteristic.

Through this, the server 100 can generate a second persona code by merging the sub-persona codes derived for each characteristic.

Through the above-described method, the method of providing a persona code according to some embodiments of the present invention generates preprocessed data using response data to the received user's questionnaire, and generates a plurality of data divided into predetermined categories through the preprocessed data. By deriving and merging each user's persona code corresponding to the characteristic, a persona code that reflects the user's hyper-personalized data can be provided.

In addition, the present invention provides a method of preprocessing unstructured data in a different form from answer data and generating a persona code using the generated preprocessed data, so that the user's preference can be more precisely reflected in the persona code.

Figure 21 is a flowchart showing a method for generating a persona code when missing data is included according to some embodiments of the present invention. Figure 22 is a block diagram illustrating a method for generating a persona code when missing data is included according to some embodiments of the present invention.

21 and 22, the preprocessing module (PM) of the server 100 may determine whether missing data exists in the data received from the user terminal 200 (S710). Specifically, the received data may include response data and unstructured data. Additionally, missing data may mean data in which the user did not respond to survey data provided to the user. That is, when the first to twelfth sub-persona codes need to be derived, data from which the third sub-persona code can be derived may not exist. At this time, the missing data may be answer data or unstructured data for deriving the third sub-persona code.

If missing data exists, the server 100 may input preprocessing data for the received data into a deep learning module that has been previously trained using training data that does not include the missing data (S720). At this time, the learning data may include answer data and unstructured data. Specifically, the learning data may include answer data and unstructured data from which all sub-persona codes can be derived for each characteristic. For example, the training data may be data from which sub-persona codes for all of the first to twelfth characteristics can be derived.

Next, the server 100 may receive at least one temporary persona code from the deep learning module (S730). Specifically, at least one temporary persona code may be received from the deep learning module as an output for the preprocessed data. At this time, the temporary persona code may be a persona code generated for received data including missing data.

Additionally, there may be multiple temporary persona codes. At this time, the plurality of temporary persona codes may be different from each other. For example, if there is missing data for the third characteristic, a sub-persona code corresponding to the third characteristic may be randomly generated. That is, a plurality of temporary persona codes may have different sub-persona codes corresponding to the third characteristic.

Meanwhile, a sub-persona code for missing data can be created using already provided training data. In other words, the sub-persona code for missing data can be generated using a pre-trained deep learning module. The learned deep learning module will be described in detail later.

Subsequently, the server 100 may derive survey data related to at least one received temporary persona code (S740). According to some embodiments, survey data may be selected from a plurality of temporary persona codes received from a deep learning module. In other words, survey data can be derived based on user response data and unstructured data with the same tendencies as temporary persona codes.

Next, the server 100 may provide a survey to the user terminal and receive response data for the provided survey data from the user terminal 200 (S750). Specifically, response data may be in the form of text, image, video, and voice for the provided survey data.

Next, the server 100 may derive the reliability of the temporary persona code based on the received response data and determine the final persona code based on this (S760). Specifically, the persona code generation deep learning module (PCDM) can send a temporary persona code to the verification module (VM). Additionally, the verification module (VM) may receive user response data.

Meanwhile, step S760 may be a step to check the accuracy of the temporary persona code. That is, step S760 can check whether the user's answer expected based on the temporary persona code matches the response data received from the user. The verification module (VM) may then generate a final persona code if the user response data and the temporary persona code do not conflict.

On the other hand, if there is a contradiction between the user response data and the temporary persona code, the verification module (VM) may modify the temporary persona code to match the user response data. Subsequently, the verification module (VM) may generate a final persona code using the modified temporary persona code.

FIG. 23 is a diagram for explaining the structure of the persona code generation deep learning module of FIG. 22. Figure 24 is a flowchart for explaining the learning method of the persona code deep learning module of Figure 22. Figure 25 is a block diagram for explaining the learning method of the persona code deep learning module of Figure 22.

23 to 25, the persona code generation deep learning module (PCDM) used in some embodiments of the present invention outputs the final persona code for the input data using an artificial neural network learned based on big data. You can.

The Persona Code Generation Deep Learning Module (PCDM) is implemented as a deep learning module and can perform artificial neural network learning using mapping data for separate parameters derived based on input data. Ewha Womans University's deep learning module can perform machine learning on parameters input as learning factors.

To explain in more detail, Deep Learning technology, a type of Machine Learning, learns at a deep level in multiple stages based on data.

Deep learning refers to a set of machine learning algorithms that extract key data from multiple data at increasing levels.

The persona code generation deep learning module (PCDM) can use a variety of known deep learning structures. For example, the persona code generation deep learning module (PCDM) can use structures such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Deep Belief Network (DBN), and Graph Neural Network (GNN).

Meanwhile, artificial neural network learning of the persona code generation deep learning module (PCDM) can be accomplished by adjusting the weight of the connection lines between nodes (adjusting the bias value if necessary) to produce the desired output for a given input. . Additionally, artificial neural networks can continuously update weight values through learning. Additionally, methods such as back propagation can be used to learn artificial neural networks.

At this time, an artificial neural network previously trained through machine learning may be installed in the memory 140 of the server 100. That is, data used for machine learning and result data may be stored in the memory 140.

In some embodiments of the present invention, the persona code generation deep learning module (PCDM) may include a transformer or autoencoder-based deep learning model including an encoder (EN) and a decoder (DE). there is. However, these are only a few examples of the present invention, and the present invention is not limited thereto.

Referring to Figure 23, the persona code generation deep learning module (PCDM) includes an input layer (input) with preprocessed data as an input node, and an output layer (output) with a persona code (or temporary persona code) as an output node. , includes M hidden layers placed between the input layer and the output layer.

Here, weights may be set on the edges connecting the nodes of each layer. The presence or absence of these weights or edges can be added, removed, or updated during the learning process. Therefore, through the learning process, the weights of nodes and edges arranged between k input nodes and i output nodes can be updated.

Before the persona code generation deep learning module (PCDM) performs learning, all nodes and edges can be set to initial values. However, when information is input cumulatively, the weights of nodes and edges change, and in this process, there is a difference between the parameters input as learning factors (i.e., preprocessed data) and the value assigned to the output node (i.e., final persona code). Matching can be achieved.

Additionally, when using a cloud server (not shown), the persona code generation deep learning module (PCDM) can receive and process a large number of parameters. Therefore, the persona code generation deep learning module (PCDM) can perform learning based on massive data.

The weights of nodes and edges between the input nodes and output nodes that make up the Persona Code Generation Deep Learning Module (PCDM) can be updated by the learning process of the Persona Code Generation Deep Learning Module (PCDM). In addition, of course, the parameters output from the persona code generation deep learning module (PCDM) can be further expanded to various data in addition to summary data.

Both semi-supervised learning and supervised learning can be used as machine learning methods used in the persona code generation deep learning module (PCDM). Additionally, the persona code generation deep learning module (PCDM) can be controlled to automatically update the artificial neural network structure to output more accurate summary data after learning according to settings.

Additionally, although not clearly shown in the drawing, in some other embodiments of the present invention, the operation of the persona code generation deep learning module (PCDM) may be performed in conjunction with the server 100 or a separate cloud server (not shown). there is.

Referring to FIGS. 24 and 25 , the server 100 may store preprocessing data for received data that does not include missing data and the final persona code derived based on the preprocessing data in a database (S810).

Here, the preprocessing data and final persona code can be derived through several embodiments of the present invention described above with reference to FIGS. 3 to 20, and are used as learning data to train the persona code generation deep learning module (PCDM). It can be. However, this is only an example of learning data of the persona code generation deep learning module (PCDM), and the present invention is not limited to this.

Next, the server 100 may load training data consisting of preprocessed data stored in the database and the final persona code for the preprocessed data (S820).

Next, the persona code generation deep learning module (PCDM) can be supervised by applying the loaded learning data to the input and output terminals of the persona code generation deep learning module (PCDM) (S830). Specifically, the server 100 generates persona codes by applying preprocessing data to the input terminal of the persona code generation deep learning module (PCDM) and applying the final persona code to the output terminal of the persona code generation deep learning module (PCDM). Deep learning module (PCDM) can be used for supervised learning. However, this is only an example, and the data received from the user terminal 200 (i.e., answer data or unstructured data) can be used as is instead of the preprocessed data applied to the persona code generation deep learning module (PCDM). Of course.

Through this, the present invention generates a plurality of temporary persona codes using a pre-trained deep learning model when the user's answer data is insufficient (i.e., when the received user data is missing data) and generates a plurality of temporary persona codes for additional users. By providing a persona code derived by verifying a temporary persona code generated using response data, the prediction accuracy of the persona code can be improved.

In addition, the present invention trains a deep learning model through supervised learning using highly accurate learning data consisting of persona codes derived based on user questionnaire response data and unstructured data, and uses the learned deep learning model to You can use this to derive and use a highly accurate persona code for missing data.

Figure 26 is a diagram for explaining the hardware configuration of a server that performs a persona code generation method according to some embodiments of the present invention.

Referring to FIG. 26, a system or server 100 that performs a persona code generation method according to some embodiments of the present invention may be implemented as an electronic device 1000. The electronic device 1000 may include a processor 1010, an input/output device 1020 (I/O), a memory 1030, an interface 1040, a storage 1050, and a bus 1060. there is. The processor 1010, input/output device 1020, memory 1030, interface 1040, and/or storage 1050 may be coupled to each other through a bus 1060. The bus 1060 corresponds to a path through which data moves.

Specifically, the processor 1010 includes a Central Processing Unit (CPU), Micro Processor Unit (MPU), Micro Controller Unit (MCU), Graphic Processing Unit (GPU), microprocessor, digital signal processor, microcontroller, and financial application processor ( It may include at least one of an AP (application processor) and logic elements that can perform similar functions.

The input/output device 1020 may include at least one of a keypad, a keyboard, a touch screen, and a display device.

The memory 1030 may load data and/or programs. At this time, the memory 1030 is an operating memory for improving the operation of the processor 1010, and may include high-speed DRAM and/or SRAM. The memory 1030 may be one or more volatile memory devices such as Double Data Rate Static DRAM (DDR SDRAM), Single Data Rate SDRAM (SDR SDRAM), and/or Electrical Erasable Programmable ROM (EEPROM), or flash memory. It may include one or more non-volatile memory devices.

The interface 1040 may perform a function of transmitting data to or receiving data from a communication network. Interface 1040 may be wired or wireless. For example, the interface 1040 may include an antenna or a wired or wireless transceiver.

The storage 1050 can store and store data and/or programs. Storage 1050 may include one or more non-volatile memory devices, such as a solid state drive (SSD), a hard drive, or flash memory. In the present invention, the storage 1050 can store a computer program consisting of instructions for performing the above-described persona code providing method.

The user terminal 200 may be a personal digital assistant (PDA), a portable computer, a web tablet, a wireless phone, a mobile phone, or a digital music player. It can be applied to music players, memory cards, or any electronic product that can transmit and/or receive information in a wireless environment.

Additionally, the server 100 and the user terminal 200 according to embodiments of the present invention may each be a system formed by connecting a plurality of electronic devices 1000 to each other through a network. In this case, each module or combination of modules may be implemented as the electronic device 1000. However, this embodiment is not limited to this.

Additionally, the server 100 is a workstation, a data center, an internet data center (IDC), a direct attached storage (DAS) system, a storage area network (SAN) system, and a network attached storage (NAS). It may be implemented as at least one of a system and a RAID (redundant array of inexpensive disks, or redundant array of independent disks) system, but the present embodiment is not limited thereto.

Additionally, the server 100 can transmit data through a network using the user terminal 200. Networks may include networks based on wired Internet technology, wireless Internet technology, and short-distance communication technology. Wired Internet technology may include, for example, at least one of a local area network (LAN) and a wide area network (WAN).

Wireless Internet technologies include, for example, Wireless LAN (WLAN), DMNA (Digital Living Network Alliance), Wibro (Wireless Broadband), Wimax (World Interoperability for Microwave Access: Wimax), and HSDPA (High Speed Downlink Packet). Access), HSUPA (High Speed Uplink Packet Access), IEEE 802.16, Long Term Evolution (LTE), LTE-A (Long Term Evolution-Advanced), Wireless Mobile Broadband Service (WMBS) and 5G NR (New Radio) technology. However, this embodiment is not limited to this.

Short-range communication technologies include, for example, Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra-Wideband (UWB), ZigBee, and Near Field Communication. At least one of NFC), Ultrasound Communication (USC), Visible Light Communication (VLC), Wi-Fi, Wi-Fi Direct, and 5G NR (New Radio) may include. However, this embodiment is not limited to this.

The server 100 that communicates through a network can comply with technical standards and standard communication methods for mobile communication. For example, standard communication methods include GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), CDMA2000 (Code Division Multi Access 2000), and EV-DO (Enhanced Voice-Data Optimized or Enhanced Voice-Data Only). , at least one of Wideband CDMA (WCDMA), High Speed Downlink Packet Access (HSDPA), High Speed Uplink Packet Access (HSUPA), Long Term Evolution (LTE), Long Term Evolution-Advanced (LTEA), and 5G New Radio (NR) may include. However, this embodiment is not limited to this.

The above description is merely an illustrative explanation of the technical idea of the present embodiment, and those skilled in the art will be able to make various modifications and variations without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are not intended to limit the technical idea of the present embodiment, but rather to explain it, and the scope of the technical idea of the present embodiment is not limited by these examples. The scope of protection of this embodiment should be interpreted in accordance with the claims below, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of rights of this embodiment.

Claims (12)

In the method of providing a persona code performed on a server linked to a user terminal,
Receiving response data to survey data pre-stored in a database linked to the server and unstructured data collected through the user terminal from the user terminal;
performing a preprocessing operation on the answer data and the unstructured data to generate preprocessing data including weights for the answer data and the unstructured data;
Generating tagged data by assigning tags to each keyword included in the preprocessed data;
generating a first persona code using the tag included in the tag assignment data;
generating a second persona code through clustering of the pre-processed data based on a category pre-specified for each survey question constituting the survey data; and
Combining the first persona code and the second persona code to generate a final persona code,
The step of generating the preprocessing data is,
Generating extracted data consisting of preset main keywords included in the response data through text mining on the response data;
determining the reliability of the answer data based on the number of updates of the answer data, the update time, and whether or not there is a contradiction in the answers of the survey data included in the answer data;
generating first preprocessing data including the extracted data and the reliability;
Extracting keywords related to objects included in the unstructured data and generating structured data for the unstructured data;
determining the importance of the structured data based on the number of updates of the unstructured data, the update time, and the proportion of the unstructured data of the part corresponding to the keyword derived from the unstructured data;
generating second preprocessing data including the standardized data and the importance level;
Deriving a first weight for the extracted data based on the first preprocessing data and deriving a second weight for the unstructured data based on the second preprocessing data;
Generating the preprocessed data including the extracted data derived based on the answer data and the first weight for the extracted data, and the structured data derived based on the unstructured data and the second weight for the extracted data; ,
The step of generating the second persona code is,
Classifying the preprocessed data according to each predetermined characteristic included in the second category of the final persona code;
performing clustering on data with high similarity among the classified data using different clustering models for each characteristic;
Deriving sub-persona codes for each cluster clustered according to the above characteristics;
Generating the second persona code by merging the plurality of sub-persona codes derived for each characteristic,
The step of performing the clustering is,
(a) calculating a silhouette coefficient for the data clustering result input to the clustering model for the first characteristic,
(b) selecting a clustering coefficient whose silhouette coefficient is greater than or equal to a preset standard value;
(c) determining clustering of data for the first characteristic based on the selected clustering coefficient;
Comprising repeating steps (a) to (c) equally for second and third characteristics different from the first characteristic.
How to provide persona codes.
According to claim 1,
The unstructured data includes at least one of image, video, voice, and location information collected in the process of receiving a user response to the survey data at the user terminal.
How to provide persona codes.
delete According to claim 1,
The step of assigning the first persona code is,
Generating first tag assigned data by assigning first tags related to a first characteristic included in the first category of the final persona code to the preprocessed data;
calculating a first score for the first characteristic by considering the weight of the preprocessed data for the keyword to which the first tag is assigned;
Generating second tagged data by assigning second tags related to second characteristics different from the first characteristics included in the first category to the preprocessed data;
calculating a second score for the second characteristic by considering a weight for the keyword to which the second tag is assigned in the preprocessed data;
assigning a first sub-persona code corresponding to the first score;
assigning a second sub-persona code corresponding to the second score;
Combining the first sub-persona code and the second sub-persona code to generate the first persona code.
How to provide persona codes.
According to clause 4,
The step of assigning the first persona code is,
Further comprising correcting the first score and the second score using user information pre-entered by the user terminal.
How to provide persona codes.
According to claim 1,
The step of assigning the first persona code is,
assigning preset tags for each topic to keywords included in the preprocessed data;
Classifying the subject tags assigned to the preprocessed data into first characteristics or second characteristics;
calculating a first score for the first characteristic by considering a weight of the preprocessed data for keywords tagged with the first characteristic;
calculating a second score for the second characteristic by considering the weight of the preprocessed data for keywords tagged with the second characteristic;
assigning a first sub-persona code corresponding to the first score;
assigning a second sub-persona code corresponding to the second score;
Combining the first sub-persona code and the second sub-persona code to generate the first persona code.
How to provide persona codes.
delete delete According to claim 1,
determining whether missing data exists in the data received from the user terminal;
If the missing data exists, inputting preprocessing data for the received data into a pre-trained deep learning module and receiving at least one temporary persona code as an output from the deep learning module;
deriving new survey data associated with the received at least one temporary persona code;
providing the survey data to the user terminal and receiving response data to the survey data from the user terminal;
Further comprising deriving reliability of the temporary persona code based on the response data and determining a final persona code based on the reliability.
How to provide persona codes.
According to clause 9,
storing the preprocessed data for the received data that does not include the missing data and the final persona code derived based on the preprocessed data in the database;
Applying the pre-processing data stored in the database to the input node of the deep learning module, and applying the persona code for the pre-processing data to the output node of the deep learning module, further comprising supervised learning of the deep learning module. containing
How to provide persona codes.
delete delete

Images