KR20230010956A - Method for determining investment indicator related with stock item and providing information of stock item based on artificial intelligence, and computing system performing the same - Google Patents

Abstract

Disclosed are a method and system for determining a reliable investment indicator for specific financial products based on artificial intelligence and providing personalized investment information to each investor based on the reliable investment indicator. According to one aspect of the present invention, a stock item information provision method comprises: a step in which a computing system constructs an item-indicator relationship DB that stores a related investment impact indicator corresponding to each of a plurality of investment items; a step in which the computing system constructs a user-indicator relationship DB that stores an interested investment impact indicator corresponding to each of a plurality of users; a step in which the computing system observes whether the related investment impact indicator corresponding to each of the plurality of investment items is updated; a step in which, when a related investment indicator is observed to be updated, the computing system determines a target user corresponding to the related investment indicator with an update among the plurality of users based on the user-indicator relationship DB; and a step in which the computing system provides the target user with personalized information about investment items corresponding to the relevant investment indicator with the update. According to the present invention, meaningful and reliable investment indicators can be determined through a plurality of documents distributed on the network.

Description

Method for determining investment indicator related with stock item and providing information of stock item based on artificial intelligence, and computing system performing the same}

The present invention relates to an artificial intelligence-based artificial intelligence-based method for determining investment indicators and providing stock information, and a computing system for performing the same. More specifically, it relates to a method and system capable of determining a reliable investment index for a specific financial product based on artificial intelligence and providing personalized investment information to each investor based on this.

If you know a keyword related to a specific financial investment product or financial investment item (eg, a specific stock item, etc.), it can be very useful, such as predicting the price change of the item through a search for information on the keyword. Therefore, in the past, attempts have been made to define and utilize related keywords for each specific financial investment item, and examples of such technologies include Korean Patent Publication No. 10-2015-0083620 (Keyword Linked Investment Information Provision System) and Korean Patent Publication No. 10 -2017-0056040 (How to provide integrated information on related items).

However, the conventional technology is a method in which a person manually maintains related keywords or related information of a specific financial investment item, or a method using exposure frequency on the web, so that it relies on a person's limited knowledge or even a keyword that is not substantially related to There was a problem with fairly low accuracy, such as common words that appear, such as the name of a specific financial product, being selected as related keywords. Therefore, a technical idea that can relatively accurately extract keywords that are significantly related to a specific financial investment item from various unstructured data circulating on the network rather than relying on the knowledge and experience of a specific person is required.

Also, in the past, there has been an attempt to predict a stock price through a plurality of unstructured data on a network such as social media. This attempt has been disclosed in Korean Patent Registration No. 10-1531970 (Stock price prediction method through analysis of social media data and stock market-related web data and stock price prediction system using the same). However, such a conventional attempt is to determine whether each word or keyword has a positive or negative effect on the stock price and collectively determine whether the corresponding document has a positive or negative effect.

However, since this method performs positive-negative judgment for each individual keyword, the entire sentence or document is classified through an appropriate algorithm (for example, Naive Bayesian) using the result of positive-negative judgment for each keyword. Even if it is classified as positive or negative, there is a problem in which errors are inevitable in practice. That is, no matter how well the positive/negative judgment for each keyword present in a sentence or document is performed, the result of determining whether the sentence or document has a positive or negative effect is inevitably error-prone.

In addition, the conventional method performs positive or negative sentiment evaluation for each keyword. In this case, it can work well for frequently used keywords (for example, public data or terms used in news) that are predetermined or formal. There is a problem in that accuracy is significantly lowered when opinions or analyzes are made on the financial product using colloquial words that users are comfortable using or keywords that are not frequently used.

Therefore, from unstructured data corresponding to specific financial products or related keywords, that is, documents, artificial intelligence-based technology that determines whether a sentence or document unit can have a positive or negative impact on the value of a financial product, rather than keywords. Technical thinking is required to design deep learning models and determine meaningful and reliable investment indicators through them.

In addition, each user's preferred investment index may be different. Therefore, each user's preferred investment index is identified in advance, and when a change in the investment index is detected, relevant investment information can be provided accurately and quickly. Technical thinking is required.

Korean Patent Publication No. 10-2015-0083620 Korean Patent Publication No. 10-2017-0056040 Korean Patent Registration No. 10-1531970

The technical problem to be achieved by the present invention is a specific financial investment item from a number of unstructured data (eg, news, posts written by users, comments, etc.; also referred to as 'unstructured documents') such as social media/internet community posts. It is to provide a technical idea that can extract practically meaningful related keywords based on artificial intelligence. Furthermore, it is to provide technical ideas that can determine meaningful and reliable investment indicators through a number of documents circulating on the network.

In addition, it is to provide technical ideas that can promptly provide information on related financial investment items to investors who are interested in specific investment indicators when changes in the relevant investment indicators occur.

According to one aspect of the present invention, the computing system, constructing an item-index relationship DB for storing the relevant investment impact index corresponding to each of a plurality of investment items; constructing, by the computing system, a user-indicator relationship DB for storing an investment impact index of interest corresponding to each of a plurality of users; Observing, by the computing system, whether an associated investment impact index corresponding to each of the plurality of investment items is updated; determining, by the computing system, a target user corresponding to the updated related investment index from among the plurality of users based on the user-index relationship DB, when a related investment index with an update is observed; and providing, by the computing system, personalized information about an investment item corresponding to the related investment index in which an update has occurred to the target user.

In an embodiment, the constructing the user-indicator relationship DB may include: collecting, for each of the plurality of users, unstructured data consumed by the user; and determining an investment impact index of interest corresponding to the user based on unstructured data consumed by the user.

In one embodiment, the method of providing item information further comprises constructing, by the computing system, an item-keyword relationship DB storing a related keyword corresponding to each of the plurality of investment items, and the user-indicator relationship. Building a DB may include: collecting unstructured data consumed by the user for each of the plurality of users; extracting a keyword of interest of the user based on unstructured data consumed by the user; extracting an investment item corresponding to the user's interest keyword from the item-keyword relation DB; and extracting an investment impact indicator corresponding to an investment item corresponding to the user's interest keyword from the item-indicator relation DB.

In one embodiment, the step of constructing an item-indicator relationship DB including related investment impact indicators corresponding to each of a plurality of investment items includes, for each of the plurality of investment items, the related investment impact indicator corresponding to the investment item. Performing a determination method, wherein the method of determining a related investment index corresponding to the investment item includes: collecting a plurality of atypical documents corresponding to the investment item or a keyword associated with the investment item; and determining an investment impact index corresponding to the investment item based on the collected atypical documents, wherein determining an investment impact index corresponding to the investment item based on the collected atypical documents comprises: Determining the investment impact index based on the output result of the impact judgment model learned through a context-sensitive natural language processing model for all or some of the collected unstructured documents to be judged; , The impact judgment model is learned to output a classification result including whether each individual document included in the documents to be judged or each sentence included in the individual document has a positive or negative impact on investment. can be a model

In one embodiment, the method for determining the related investment index corresponding to the investment item further includes performing filtering on unstructured documents corresponding to a predetermined filtering condition among collected unstructured documents, and after performing the filtering It may be characterized in that the remaining atypical documents are specified as the documents to be judged.

In one embodiment, the step of filtering on unstructured documents corresponding to a predetermined filtering condition among the unstructured documents collected by the system includes generating a document vector for each of the collected unstructured documents. step; clustering atypical documents having a similarity of at least a certain level based on the generated document vectors; and performing filtering to exclude a part of each of the clustered irregular document clusters from the document to be judged.

In one embodiment, the clustering of irregular documents having a similarity of at least a predetermined level based on the generated document vectors includes clustering between irregular documents having a similarity of at least a predetermined level and a generation time of the unstructured documents within a predetermined time range. It may include the step of performing.

In one embodiment, the step of collecting a plurality of unstructured documents corresponding to the investment item or a keyword related to the investment item includes the first unstructured documents corresponding to the investment item and the second unstructured document corresponding to the related keyword. The step of determining an investment impact index corresponding to the investment item based on the collected atypical documents includes a first investment impact index extracted based on the first atypical documents and and determining based on a second investment impact index extracted based on the second atypical documents.

In one embodiment, the step of determining based on the first investment impact index extracted based on the first unstructured documents and the second investment impact index extracted based on the second unstructured documents, It may be characterized in that the investment impact index is determined to have different weights for each impact index and the second investment impact index.

In one embodiment, the impact determination model outputs a classification result including whether each sentence included in the individual document has a positive or negative impact on investment, and based on the classification result for each sentence, It may be characterized in that the classification result of the individual document is judged.

According to another aspect of the present invention, a computer program stored in a computer readable recording medium is provided to perform the above method.

According to another aspect of the present invention, a computer readable recording medium on which a computer program for performing the above method is recorded is provided.

According to another aspect of the present invention, a computing system comprising: a processor; and a memory for storing a computer program executed by the processor, wherein the computer program, when executed by the processor, causes the computing system to store a related investment impact index corresponding to each of a plurality of investment items. constructing an item-indicator relation DB; Constructing a user-indicator relationship DB that stores investment impact indicators of interest corresponding to each of a plurality of users; Observing whether the related investment impact index corresponding to each of the plurality of investment items is updated; determining a target user corresponding to the related investment index in which an update has occurred among the plurality of users based on the user-indicator relationship DB when a related investment index in which an update has occurred is observed; and providing the target user with personalized information about an investment item corresponding to the related investment index in which an update has occurred.

According to an embodiment of the present invention, it is possible to extract related keywords that are substantially meaningful to a specific financial investment event based on artificial intelligence from a plurality of unstructured data such as social media/internet community posts.

In addition, the effect of extracting keywords that are substantially related to the financial investment item by considering the meaning of the information from live information at the time of keyword extraction, rather than simply extracting related keywords according to the fragmentary knowledge of a specific person or exposure frequency there is

In addition, due to the nature of unstructured data such as social media, when the same or almost similar content is uploaded on the network by different publishers (for example, the same or very similar news article is uploaded on the network by multiple media, or a specific person's When content is repeatedly uploaded by a large number of users) is frequent, there is an effect of improving accuracy by extracting related keywords after filtering is performed to a certain extent to eliminate redundancy of such unstructured data.

In addition, when expanding to other keywords that are related to the related keyword, it is possible to search not only financial investment items that are directly related to the issue, but also financial investment items that are indirectly related to the issue.

In addition, through a large number of documents circulating on the network, a deep learning model can be designed to determine whether the document is positive or negative for a specific financial product in a sentence or document unit, and through this, a meaningful and reliable related investment index can be determined. There is an effect.

In addition, based on these related investment indicators, personalized information on specific investment indicators can be provided. In other words, when a change in a specific investment indicator is detected, information on an investment item closely related thereto can be quickly provided to investors who are interested in the investment indicator.

In order to more fully understand the drawings cited in the detailed description of the present invention, a brief description of each drawing is provided.
1 is a diagram for explaining schematic system configurations for performing an AI-based personalized event information providing method according to the technical idea of the present invention.
2 is a block diagram showing a schematic and logical configuration of an item information providing system according to an embodiment of the present invention.
3 is a diagram for explaining a schematic physical configuration of an event information providing system according to an embodiment of the present invention.
4 is a block diagram for explaining a schematic configuration of an item DB construction module that performs a method for determining a keyword related to a specific item/related investment impact index according to an embodiment of the present invention.
5 is a diagram for explaining the concept of a natural language processing model according to an embodiment of the present invention.
6 is a diagram for explaining the concept of extracting a related keyword through a word vector according to an embodiment of the present invention.
7 is a diagram for explaining a concept of performing filtering according to an embodiment of the present invention.
8 is a diagram for explaining changes in related keywords according to an embodiment of the present invention.
9 is an exemplary flowchart showing a process of constructing an item-keyword relation DB according to an embodiment of the present invention.
10 is a diagram for explaining a process of constructing an event-indicator relation DB according to an embodiment of the present invention.
11 is a diagram for explaining the concept of determining an investment impact index using atypical documents according to an embodiment of the present invention.
12 is a diagram for explaining the concept of performing the impact evaluation of a document through sentiment evaluation in sentence units according to an embodiment of the present invention.
13 is a diagram for explaining the concept of adaptively performing impact evaluation by reflecting the order of a document according to an embodiment of the present invention.
14 is a flowchart illustrating an example of a method for providing personalized item information based on artificial intelligence according to an embodiment of the present invention.

Since the present invention can apply various transformations and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, it should be understood that this is not intended to limit the present invention to specific embodiments, and includes all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the present invention, if it is determined that a detailed description of related known technologies may obscure the gist of the present invention, the detailed description will be omitted.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. These terms are only used for the purpose of distinguishing one component from another.

Terms used in this application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this specification, terms such as "include" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other It should be understood that the presence or addition of features, numbers, steps, operations, components, parts, or combinations thereof is not precluded.

In addition, in the present specification, when one component 'transmits' data to another component, the component may directly transmit the data to the other component, or through at least one other component. It means that the data can be transmitted to the other component. Conversely, when one component 'directly transmits' data to another component, it means that the data is transmitted from the component to the other component without going through the other component.

1 is a diagram for explaining schematic system configurations for performing an AI-based personalized event information providing method according to the technical idea of the present invention.

Referring to FIG. 1, an artificial intelligence-based personalized item information providing method according to an embodiment of the present invention is performed by a predetermined financial investment item information providing system (100; hereinafter referred to as 'item information providing system'). can

The item information providing system 10 determines a financial investment item (hereinafter referred to as 'issue item') related to the issue when various social issues occur, and the user terminal ( 30) can be provided. A financial investment item may mean, for example, an individual item of a financial product such as stocks, futures, and options.

In particular, the item information providing system 10 is an investment indicator (also referred to as an 'investment factor' that the user of the user terminal 30 is usually interested in, and is also referred to as an 'investment impact indicator' because it affects the stock price of a related item. ) can be analyzed, extracted, and databased in advance, and when an update to the investment index of interest occurs, information on related financial investment items related to the investment index of interest, that is, personalized information to the user can be provided.

Information on a specific financial investment item provided to users includes the name of the item, the theme, the flow of stock prices (eg, a time series chart such as a bar chart), news related to the item, analyst analysis reports related to the item, and the corresponding item. Market information on sectors included, disclosure information, various investment indicators related to the stock (e.g., market capitalization, PER, PBR, dividend yield, 52-week new price, 52-week new low, moving average, etc.), Information on the included investment strategy, investment logic, and portfolio may be included, but is not limited thereto, and may include various information or contents related to the financial investment item.

As will be described later, the item information providing system 10 may use unstructured data to identify investment indicators of interest of individual users, keywords related to a specific item, or investment indicators related to a specific item, and the item information providing system (10) can collect various unstructured data through the network. In this specification, unstructured data may refer to various contents circulating on a network, and such unstructured data may be generated by media companies, financial institutions, securities companies, and general users. The unstructured data can be treated as a kind of document, and if text is included, the length of the content is long (eg, several lines or more than several tens of lines) or short (eg, one line of comments, etc.) It can be the unstructured data regardless. In addition, the unstructured data defined in this specification means that the data collected by the item information providing system 10 is not limited to a predetermined format, but each producer of the unstructured data has a certain format and rules. An average expert in the field of the present invention will be able to easily infer that data can be generated.

The item information providing system 10 may acquire such unstructured data from a predetermined information source system 20 . The information source system 20 may be a system capable of generating unstructured data or storing and registering/posting unstructured data generated by a producer (ie, a user, manager, or publisher of the information source system 20). . The information source system 20 includes, for example, a social media service system, various online community service systems, a media company system, a financial institution system, a stock company system, and a function for users to post posts or comments in text form. A variety of systems may be provided.

Although FIG. 1 shows only one representative information source system 20 for convenience of explanation, in reality, the item information providing system 10 can collect unstructured data from numerous information source systems existing on the network.

The item information providing system 10 may crawl various unstructured data periodically or according to a manager's command. Depending on implementation, unstructured data may be collected through a predetermined protocol (eg, API) in a system to which the unstructured data is collected (ie, the information source system 20). According to an example, the item information providing system 10 may collect a large amount of unstructured data in bulk and then extract and use only necessary specific unstructured data, and may use specific conditions (eg, specific user, specific item, specific period). It is also possible to collect only unstructured data that satisfies .

On the other hand, the item information providing system 10 may be a computing system that is a data processing device having an arithmetic capability for implementing the technical idea of the present invention, and generally not only a server that is a data processing device accessible to clients through a network. It may include a computing device such as a personal computer or a portable terminal.

The item information providing system 10 may be implemented with any one physical device, but if necessary, a plurality of physical devices may be organically combined to implement the item information providing system 10 according to the technical idea of the present invention. An average expert in the art of the present invention will be able to easily infer that there is.

The item information providing system 10 may be implemented in the form of a subsystem of a predetermined parent system (not shown). The parent system may be, for example, a securities company system, a system of a financial institution such as a bank, or a system that provides community services or various types of investment information. Also, the parent system may be a server. The server refers to a data processing device having computing power for implementing the technical idea of the present invention, and generally can perform a specific service, such as a personal computer or mobile terminal, as well as a data processing device accessible to clients through a network. An average expert in the art of the present invention can easily infer that any device that exists can be defined as a server.

The user terminal 30 may be a computing device having a processing device. For example, the user terminal 200 may be a mobile device such as a smart phone, a mobile phone, a PDA, or a tablet PC, but is not limited thereto, and may include a laptop, a desktop, and the like. Meanwhile, the user terminal 200 may include a microphone capable of receiving a speaker's voice, a display device capable of outputting contents for learning provided by the learning system 100, and a speaker.

Meanwhile, the item information providing system 10 is connected to the user terminal 20 through a wired/wireless network (eg, Internet) to provide various information, data, and/or signals necessary to implement the technical idea of the present invention. can transmit and receive.

2 is a block diagram showing a schematic and logical configuration of an item information providing system 10 according to an embodiment of the present invention, and FIG. 3 is a schematic block diagram of the item information providing system 10 according to an embodiment of the present invention. It is a drawing for explaining the physical configuration of

Referring to FIG. 2, the item information providing system 10 according to an embodiment of the present invention includes a data collection module 50, an item DB construction module 100, a user-index relationship DB construction module 300, and an update judgment module. 300, an information providing module 400, and a DB 500 may be included. Depending on the embodiment of the present invention, some of the above-described components may not necessarily correspond to components essential to the implementation of the present invention, and also according to the embodiment, the item information providing system 10 Of course, may include more components than this. For example, the item information providing system 10 includes a communication module (not shown) for communicating with an external device, a control module (not shown) for controlling components and resources of the item information providing system 10, various A storage module (not shown) capable of storing data and information may be further included.

The item information providing system 10 may refer to a logical configuration having hardware resources and/or software required to implement the technical idea of the present invention, and necessarily refers to one physical component or It does not mean a single device. That is, the item information providing system 10 may mean a logical combination of hardware and/or software provided to implement the technical idea of the present invention, and if necessary, it is installed in devices spaced apart from each other to perform respective functions. By performing, it may be implemented as a set of logical configurations for implementing the technical idea of the present invention. In addition, the item information providing system 10 may mean a set of components implemented separately for each function or role to implement the technical idea of the present invention. For example, the data collection module 50, item DB construction module 100, user-index relationship DB construction module 200, update judgment module 300, information provision module 400, DB 500, respectively may be located on different physical devices or may be located on the same physical device. In addition, depending on the implementation example, the data collection module 50, item DB construction module 100, user-index relationship DB construction module 200, update judgment module 300, information provision module 400, DB ( 500) A combination of software and/or hardware constituting each may also be located in different physical devices, and components located in different physical devices may be organically combined with each other to implement each of the modules.

Also, in this specification, a module may mean a functional and structural combination of hardware for implementing the technical concept of the present invention and software for driving the hardware. For example, the module may mean a logical unit of a predetermined code and a hardware resource for executing the predetermined code, and necessarily means a physically connected code or one type of hardware. That this is not the case can be easily deduced to the average expert in the art of the present invention.

Meanwhile, the item information providing system 10 may physically have a configuration as shown in FIG. 3 . The item information providing system 10 includes a memory 120-1 storing a program for implementing the technical idea of the present invention, and a processor 110-1 for executing the program stored in the memory 120. It can be.

An average expert in the technical field of the present invention can easily infer that the processor 110-1 can be named by various names such as CPU, APU, mobile processor, etc., depending on the implementation example of the item information providing system 10. You will be able to. In addition, as described in FIG. 2, the item information providing system 10 may be implemented by organically combining a plurality of physical devices. In this case, at least one processor 110-1 is provided for each physical device. An average expert in the technical field of the present invention will be able to easily infer that the item information providing system 10 of the present invention can be implemented.

The memory 120-1 stores the program and may be implemented as any type of storage device accessible by the processor to drive the program. Also, according to a hardware implementation example, the memory 120-1 may be implemented as a plurality of storage devices rather than a single storage device. In addition, the memory 120-1 may include a temporary storage device as well as a main memory device. In addition, it may be implemented as a volatile memory or a non-volatile memory, and may be defined as including all types of information storage means implemented so that the program can be stored and driven by the processor.

According to an embodiment, the item information providing system 10 may generate personalized information on items closely related to an investment index of interest according to the technical idea of the present invention, and a subject who wants to use it (eg, a securities company) , banks, or other service providers, etc.), can be implemented in various ways such as web servers, computers, etc., and any type of data processing device capable of performing the functions defined in this specification It can be defined in a meaning that includes.

In addition, according to the embodiment of the event information providing system 10, various peripheral devices (peripheral device 1 to peripheral device N, 130-1, 131-1) may be further provided. For example, an average expert in the technical field of the present invention may easily infer that a keyboard, monitor, graphic card, communication device, etc. may be further included in the item information providing system 10 as peripheral devices.

Hereinafter, when a predetermined module performs a function in this specification, it means that the processor 110-1 drives a program included in the memory 120-1 to perform the function. An average expert in the art will be able to reason easily.

Referring to FIG. 2 , the DB 500 may store various data or DB tables required to implement the technical idea of the present invention. For example, in the process of performing the AI-based personalized item information providing method according to the technical concept of the present invention, the item information providing system 10 includes an item-keyword relationship DB, an item-indicator relationship DB, and a user- An indicator relation DB can be constructed. In addition, the constructed item-keyword relationship DB, item-index relationship DB, and user-index relationship DB may be stored in the DB 500 .

In this specification, DB may be implemented as at least one table, and may be used to mean further including a separate database management system (DBMS) for searching, storing, and managing information stored in the DB. In addition, it can be implemented in various ways such as linked-list, tree, relational DB, etc., and includes all data storage media and data structures capable of storing information to be stored in the DB. can be used

Meanwhile, the data collection module 50 may collect a plurality of unstructured data. To this end, the data collection module 50 may crawl web or app information provided by the information source system 200 or collect it through an API provided by the system. Such an information source system may be various, such as a Social Network Service (SNS) system, a media company, an exchange, a financial institution, a portal, a cafe, and a blog.

The data collection module 50 may collect a large amount of unstructured data in bulk and then extract and use only necessary specific unstructured data, and unstructured data that satisfies specific conditions (eg, specific user, specific item, specific period). You can only collect them.

Meanwhile, the item DB construction module 100 may construct an item-keyword relation DB for storing related keywords corresponding to each of a plurality of investment items. In addition, the item DB construction module may construct an item-indicator relation DB storing related investment impact indicators corresponding to each of a plurality of investment items.

For each of a plurality of financial investment items, the item DB construction module 100 may extract a related keyword corresponding to the financial investment item by performing a method of determining a related keyword corresponding to the investment item, and each financial investment item The item-keyword relation DB may be constructed by storing or updating related keywords corresponding to the item in the item-keyword relation DB.

In addition, for each of a plurality of financial investment items, the item DB construction module 100 may extract related investment impact indicators corresponding to the investment items, and convert related investment indicators corresponding to each financial investment item into item-indices. The item-indicator relation DB can be constructed by storing or updating in the relation DB. To this end, the item DB construction module 100 may determine an investment index capable of predicting the value change of a specific financial item using unstructured data (documents) collected from networks such as social media according to the technical idea of the present invention. In particular, in the case of determining the investment impact index of a specific financial item using keywords related to a specific financial item determined according to the technical idea of the present invention, it is more reliable and reflects social awareness (changes in related keywords, etc.) that change over time. There is an effect that can determine the investment impact index.

Financial investment items refer to individual items of financial instruments such as stocks, futures, and options, and in this specification, financial items or investment items may be used as the same meaning as financial investment items. The plurality of financial investment items may be predefined.

A related keyword corresponding to a specific financial item is, theoretically, a keyword representing an event, object, person, trend, etc. that affects the value change of the specific financial item (eg, stock item A, bond B, etc.). can mean Substantially, according to the technical concept of the present invention, it is considered to be related to a specific financial item based on a plurality of unstructured data (eg, user content on social media such as SNS, news, content disclosed in cafes or blogs, etc.) It may mean an extracted keyword.

The item DB construction module 100 may collect unstructured data corresponding to the specific financial item through a network. The item DB construction module 100 may collect unstructured data corresponding to the specific financial item from the information source system 200, for example.

The unstructured data corresponding to the specific financial item may refer to various contents that include the name of the specific financial item (for example, Samsung Electronics, Hynix, etc.) and circulate on the network. , can be created by financial institutions, securities companies, and general users. The unstructured data can be treated as a kind of document, and if text is included, the length of the content is long (eg, several lines or more than several tens of lines) or short (eg, one line of comments, etc.) It can be the unstructured data regardless.

The item DB construction module 100 may crawl unstructured data corresponding to a specific financial item periodically or according to a manager's command. Depending on implementation, the unstructured data may be collected through a predetermined protocol (eg, API) in a system to which the unstructured data is to be collected. According to an example, after a large amount of unstructured data is collected regardless of whether it is unstructured data corresponding to the specific financial item, only data corresponding to the specific financial item may be extracted by the item DB building module 100, According to another example, only unstructured data corresponding to the specific financial item may be collected through a network.

Then, the item DB construction module 100 may determine a keyword related to the specific financial item based on the unstructured data collected in this way.

That is, according to the technical concept of the present invention, the item DB construction module 100 is a related keyword of the specific financial item based on opinions, reactions, and analyzes written by a plurality of content producers currently circulating on the network at the time of determining the relevant keyword. There are characteristics that determine To this end, the item DB building module 100 may limit unstructured data used to determine keywords related to a specific financial item based on the production time of the unstructured data (eg, upload time on the network). . For example, the item DB building module 100 may determine a keyword related to a specific financial item based only on unstructured data generated in a recent predetermined period (eg, 1 month, 3 months, etc.).

In addition, by periodically performing the determination of such related keywords and monitoring changes in related keywords of the specific financial issue, changes in the business direction of the specific financial issue or the subject of the specific financial issue (eg, company, etc.) or a meaningful event There is an effect of confirming the occurrence of based on related keywords.

According to the technical idea of the present invention, the item DB construction module 100 determines keywords associated with a specific financial item by analyzing unstructured data circulating on the network, and analysis of such unstructured data includes a deep learning-based natural language processing model ( Natural Language Processing Model) can be used.

In particular, in order to determine related keywords with higher accuracy, the item DB building module 100 may use a context sensitive natural language processing model.

A context-sensitive natural language processing model is a concept opposite to a context-free natural language processing model, and may refer to a natural language processing model that differently defines or recognizes the meaning of a certain keyword or sentence according to context. In contrast, a context-free natural language processing model may mean defining or recognizing the meaning of a corresponding keyword or sentence as a corresponding keyword or sentence itself regardless of context.

This context-sensitive natural language processing model learns a large number of learning target data and actually uses a specific keyword (sometimes referred to as a token in a natural language processing model, and a keyword can be a single token or a combination of tokens. In this specification, for convenience of explanation, One of the important performance of the natural language processing model is to understand the keywords related to the token or the combination of tokens as a keyword), and for this purpose, a natural language processing model applying the concept of attention has appeared. did

Since attention is related to a keyword in order to define or recognize a certain keyword, it can mean a relationship with other keywords when referred to with more attention. For example, a second keyword having a high value of attention with the first keyword may mean that it is used in learning data with a high correlation with the specific keyword.

Since the concept of attention, such as the attention mechanism and the attention function, is widely known, a detailed description thereof will be omitted.

In the case of a natural language processing model that vectorizes keywords used in training data using the concept of attention, each keyword included in training data can be vectorized by reflecting the context.

Examples of such natural language processing models include, for example, Embeddings from Language Model (ELMo), Universal Language Model Fine-tuning for Text Classification (ULM-FiT), and Bidirectional Encoder Representations from Transformers (BERT).

In any case, the natural language processing model performs a task of vectorizing a keyword (word embedding), and through vectorization of such a keyword, the meaning of the keyword can be expressed as a vector.

In particular, in the case of BERT (Bidirectional Encoder Representations from Transformers), a large amount of training data can be learned without a labeling task through unsupervised learning, and the keyword can be vectorized with high performance differentially according to the context of each keyword. It is getting attention as a model that can do it.

Therefore, the natural language processing model according to an embodiment of the present invention may be a salmon fish processing model that vectorizes each keyword by learning learning data through the BERT natural language processing model, but is not limited thereto.

In any case, according to the technical idea of the present invention, when the vectorization result of each keyword reflecting the context is used through a context-sensitive natural language processing model, it is much higher than the conventional method of determining related keywords simply through statistical frequency, etc. You can have a performance (that is, words in a general sense that simply appear like a specific keyword with a high frequency but have little practical relevance).

Meanwhile, according to the technical idea of the present invention, the item DB construction module 100 may determine an investment impact index of a specific financial item based on unstructured data on the network. At this time, although keywords related to specific financial items determined as described above may be used, it is not necessarily limited thereto, and investment impact indicators according to the technical idea of the present invention may be used even when keywords related to specific financial items determined by experts or the like are used. can also decide

To this end, the item DB construction module 100 may use a deep learning-based natural language processing model. The natural language processing model for determining the investment impact index may be provided separately from the natural language processing model for determining the related keyword.

Such a natural language processing model may be named an influence judgment module as will be described later. The natural language processing model can determine whether unstructured data, that is, sentences included in collected documents, have a positive or negative effect on a particular financial item, or determine whether the documents themselves have a positive or negative effect. It may be a deep learning model trained to Hereinafter, in the present specification, a natural language processing model learned to determine a related keyword will be referred to as a first natural language processing model. In addition, the natural language processing model for determining the investment impact index, that is, the impact judgment model, will be named a second natural language processing model.

The first natural language processing model and the second natural language processing model may be separately provided. According to an embodiment, the second natural language processing model establishes a first natural language processing model and then fine-tunes the first natural language processing model to determine whether the sentence or document is positive when a specific sentence or document is input. It can be a model trained as a classification model that classifies whether it is negative or neutral.

In this way, a machine learning model that outputs a classification result including positive or negative when specific data is input is called a Sentiment Analysis model.

According to an embodiment, the second natural language processing model may be a model that classifies input data into three classes of positive, negative, or neutral, and according to other embodiments, strong positive, weak positive, strong negative, weak negative, It can also be a model that is trained to classify the input data into more classes, such as neutral.

This second natural language processing model pre-trains a context-sensitive natural language processing model (eg, BERT, etc.) with a large amount of corpus (or documents), and then uses the pre-trained model as described above. It can be built by fine-tuning to perform the same sentiment analysis. As described above, in this pre-learning, the first natural language processing model is built by additionally learning data related to the open natural language processing model itself or financial products such as BERT, and the first natural language processing model is used as a plurality of labeled training data It can be built by fine-tuning using . The plurality of labeled training data may be data in which classification results (eg, positive, negative, or neutral) to be output by the second natural language processing model for a plurality of sentences are labeled. Of course, when the document itself is used as training data, the data labeled on the document itself may be the training data.

In any case, the second natural language processing model according to the technical concept of the present invention may output a classification result including whether the corresponding sentence is positive or negative for a specific financial item for each sentence. This can lead to a judgment result with much higher accuracy and reliability than conventionally performing positive or negative sentiment analysis for each keyword in a predetermined method.

That is, in the case of sentiment analysis by keyword, which is a conventional method, when positive keywords and negative keywords coexist in a sentence, the probability that the sentence or document is positive or negative is determined through a probabilistic model (eg, the naive Bayesian method, etc.) In this case, there was a problem that the accuracy of sentiment analysis was relatively low.

However, in the case of a deep learning model that performs sentiment analysis on a sentence or document unit itself using a context-sensitive natural language processing model learned from a large amount of data, a sentence or document whose meaning itself is similar to the labeled training data is compared with the corresponding labeling result. It is learned to produce the same output, and in this case, there is an effect that much more reliable sentiment analysis can be performed.

Hereinafter, in the present specification, a case in which the second natural language processing model performs sentiment analysis for each sentence included in the collected unstructured documents and through this is described as an example, but depending on the embodiment, the document itself An average expert in the technical field of the present invention can easily infer that the second natural language processing model can be learned and utilized to output a sentiment analysis result for .

On the other hand, the technical concept of the present invention has the advantage of being able to reflect the temporal trend while reflecting various opinions of a number of content producers by determining the related keyword/related investment impact index of a specific financial item through unstructured data on the network. In addition, technical ideas that can solve problems that may occur when using unstructured data collected on the network can be provided together.

A typical problem that may occur when using unstructured data collected on a network may be information redundancy. That is, there is a problem in that the same or substantially the same content is produced as separate unstructured data by a plurality of content producers. In this case, even if it is actually one content (text document), if it is repeatedly uploaded by multiple producers or uploaded with only slight changes applied, the problem of repeatedly learning the one content by the natural language processing model may occur. there is.

In addition, this may have a problem in that the natural language processing model is overfitted to the corresponding content, that is, it may have a tendency to be learned to have a high degree of correlation even though there is no actual correlation or a low degree of correlation.

To this end, the item DB construction module 100 may filter unstructured data through a predetermined criterion to provide a technical concept for resolving redundancy.

After all, according to the technical idea of the present invention, a context-sensitive natural language that can understand the meaning of related keywords of a specific financial item to some extent from a plurality of unstructured data on the network, that is, a plurality of live information produced by a plurality of producers. It is determined with high accuracy through the processing model, but there is an effect of solving problems that may occur when using a large number of unstructured data on the network. Also, based on a plurality of live information produced by a plurality of producers, it is possible to determine a predetermined investment impact index capable of predicting the stock price direction of a specific financial issue.

The detailed configuration of the item DB building module 100 for implementing these technical ideas will be described with reference to FIG. 4 .

4 is a block diagram for explaining a schematic configuration of an item DB construction module 100 that performs a method for determining a keyword related to a specific item/related investment impact indicator according to an embodiment of the present invention.

Referring to FIG. 4 , the item DB construction module 100 according to the technical idea of the present invention may include an extraction module 110 and a natural language processing model 120 (including an influence judgment model). According to an embodiment, the item DB building module 100 may further include a filtering module 130 and a document vector generating module 140.

The data collection module 50 may collect a plurality of unstructured data corresponding to a specific financial item. Then, the extraction module 110 determines the specific data based on the collected unstructured data (eg, financial news, public announcements, contents uploaded by users, various document contents such as analyst reports, or comments thereon). A related keyword corresponding to a financial item may be extracted, and a related keyword corresponding to the specific financial item may be stored in an item-keyword relation DB.

To this end, the extraction module 110 may learn target data that are all or part of unstructured data collected through the natural language processing model 120 . As described above, the natural language processing model 130 may include a first natural language processing model and a second natural language processing model. As described above, the second natural language processing model outputs classification results including whether sentences or documents included in unstructured data, that is, unstructured documents, have a positive or negative effect on the specific financial item as described above. It may be an influence judgment model or a sentiment analysis model that

As described above, the first natural language processing model included in the natural language processing model 120 may be a model capable of vectorizing each of the keywords included in the learning target data in a context-sensitive manner.

For example, the first natural language processing model may be a model that can learn a large amount of corpus through unsupervised learning, such as BERT, and train only the learning target data or other training target data. It can be learned in such a way that a large amount of documents, that is, corpuses are learned in advance and then the learning target data are additionally trained.

For example, FIG. 5 is a diagram for explaining the concept of a natural language processing model according to an embodiment of the present invention. As shown in FIG. 5, a predetermined first natural language processing model (eg, an NLP model such as BERT) ) may be provided.

The first natural language processing model performs learning on learning target data (eg, D1, D2, D3, D4, etc.).

The first natural language processing model may be trained with only the learning target data when there is a sufficiently large amount of target data. It may be a pre-trained model with other large amounts of data (for example, wiki data in the case of BERT) other than learning target data.

In addition, word vectors for each of the keywords included in the learning target data may be obtained by additionally learning the learning target data.

The learning target data may be data remaining after resolving a redundancy problem among unstructured data corresponding to a specific financial item collected by the data collection module 50 . The technical idea for solving the redundancy problem will be described later, and if it is determined that different unstructured data have redundancy through a predetermined criterion, the learning target data can be specified by leaving only one of them (or several according to the embodiment). there is.

In any case, the first natural language processing model may be a model capable of vectorizing each of the keywords included in the learning target data by reflecting the context.

Then, the extraction module 110 obtains a first vector corresponding to the keyword of the specific financial item (ie, the name of the specific financial item) acquired through the learned natural language processing model 120 and the first vector and the predetermined At least one second vector satisfying the criterion may be extracted, and a keyword corresponding to the extracted at least one second vector may be extracted as the related keyword.

A natural language processing model with excellent performance, that is, a well-trained natural language processing model, can vectorize keywords so that keywords having the same or similar meanings can exist in close positions on a vector space.

And these vectorized keywords can be mapped on a vector space (eg, 768-dimensional space in the case of BERT).

This concept will be described with reference to FIG. 6 .

6 is a diagram for explaining the concept of extracting a related keyword through a word vector according to an embodiment of the present invention.

Referring to FIG. 6 , each of the keywords included in the learning target data may be vectorized through the learned first natural language processing model, and this may be shown in FIG. 6 .

In FIG. 6, the first vector 1 may be a vector representing a keyword (name) corresponding to a specific financial item.

In addition, each vector (eg, 2-1 to 2-8) of the keywords learned through the first natural language processing model, which is closely related to the keyword, is mapped to a close position on the vector space.

At this time, the first natural language processing model performs word embedding (keyword vectorization) in consideration of the context, and thus, in many cases, it is simply described with a keyword corresponding to a specific financial item, but when it is described together with other financial items, substantial Words that are not semantically related (eg, keywords commonly used in stocks, verbs such as 'is', etc.) are distanced from the first vector (1) corresponding to the specific financial item in the vector space. may be mapped relatively far.

Accordingly, the extraction module 110 may extract keywords related to the specific financial item through the result of vectorizing the keywords included in the learning target data by the first natural language processing model.

For example, vectors in which the first vector 1 and a predetermined predetermined criterion, for example, a cosine similarity greater than or equal to a predetermined value or a Euclidean distance between the first vector and the first vector 1 are less than or equal to a predetermined value (For example, 2-1 to 2-8) may be extracted, and keywords corresponding to the extracted vectors may be extracted as related keywords.

An average expert in the field of the present invention can easily infer that cosine similarity or Euclidean distance are both metrics that can be used to measure the similarity of vectors in a vector space.

In addition, it is of course possible to determine how much the predetermined value is determined through an experiment.

Although the related keywords extracted in this way may be further restricted by a service manager or additional processing, in any case, the related keywords extracted in this way may be specified as the related keywords of the specific financial item or their candidates.

On the other hand, the item DB building module 100 does not learn all unstructured data corresponding to a specific financial item as learning target data, but performs a predetermined filtering and specifies only unstructured data remaining as a result of the execution as learning target data. .

To this end, the filtering module 130 may perform filtering on unstructured data corresponding to a predetermined filtering condition among the unstructured data collected by the data collection module 50 . In addition, unstructured data remaining after filtering may be specified as learning target data.

According to an example, the filtering module 130 may filter out unstructured data with high redundancy among the collected unstructured data and specify only those with low redundancy among learning target data.

To this end, the filtering module 130 may cluster or group unstructured data, that is, documents having a similarity higher than a certain level.

To this end, the item DB construction module 100 may create a document vector representing each unstructured data and utilize it. The document vector is a vector for characterizing the corresponding document and is a sentence including keywords corresponding to the corresponding word vector using word vectors obtained through the natural language processing model 120 where word embedding is well performed in consideration of the context. Of course, a method of defining a sentence vector characterizing a sentence vector or a document vector characterizing a corresponding document including corresponding sentences may be varied.

Of course, it is not necessary to use the word vector generated by the natural language processing model 120 to generate a document vector, and various known document vectors (eg, TF-IDF or An average expert in the field of the present invention can easily infer that the production method can be used.

Generation of such a document vector may be performed by the document vector generation module 140 .

According to an example, the document vector generation module 140 may generate a document vector by utilizing a word vector of each of keywords included in a document obtained through the natural language processing model 120 . A sentence vector can be created using a word vector and a document vector can be created using it, or a document vector can be defined directly from a word vector. In order to derive a sentence vector and/or a document vector, each word vector of each keyword included in a sentence or document may be calculated in a predetermined method, and each method may vary according to embodiments, of course. In particular, in the case of a model that performs word embedding in consideration of deep attention in both directions, such as BERT, the performance of sentence vectors or document vectors generated through word vectors can be improved.

Then, the filtering module 130 may cluster unstructured data whose similarity is equal to or higher than a certain level based on the document vectors for each of the unstructured data generated by the document vector generation module 140 . Cosine similarity or Euclidean distance as described above may also be used for the similarity between document vectors, and various embodiments may be possible for how to determine the criterion value for clustering.

Then, the filtering module 130 may perform filtering to exclude some of the clustered unstructured data clusters from the learning target data.

Meanwhile, according to the technical idea of the present invention, when clustering is performed, clustering may be performed by further considering the generation time (eg, upload time, etc.) of each of the unstructured data.

In general, when a certain content is created, duplicate content based on the content (for example, content uploaded by different publishers of the same or similar content) is frequently created at a time similar to the time of creation of the content. In this case, it may be desirable to filter duplicate content. However, if there is a difference of more than a certain period of time even if the contents are similar, this is because it can be recognized and handled by a number of users as its own content.

Therefore, even if the content has a similar meaning, i.e. unstructured data, if the time interval at the time of creation is large, it may be preferable to treat it as new content that can newly affect the market or affect user perception. . Therefore, the technical idea of the present invention can perform filtering on unstructured data by accommodating this concept.

An example of this may be as shown in FIG. 7 .

7 is a diagram for explaining a concept of performing filtering according to an embodiment of the present invention.

Referring to FIG. 7 , document vectors (eg, D1 to D19) corresponding to each of unstructured data may be generated from a network by the document vector generation module 140 .

Each of the document vectors (for example, D1 to D19) shown in FIG. 7 may be document vectors clustered into one cluster according to the degree of similarity between the document vectors.

According to an embodiment, since the document vectors classified into any one cluster (for example, D1 to D19) are actually unstructured data with similar contents, one document or several documents with the lowest similarity as needed It is possible to specify the learning target data by leaving only the document, that is, unstructured data, as the learning target data and performing filtering on the rest. And this process can be performed for each cluster.

In addition, according to embodiments, the filtering module 130 may perform clustering based not only on the similarity of documents but also on the creation time of the corresponding document. That is, clustering may be performed between documents that satisfy the condition that the similarity of the documents is at least a certain level and the document creation time is within a certain interval, and the result may be as shown in FIG. 7 .

That is, document vectors (for example, D1 to D19) classified as one cluster based only on document similarity can be classified into three sub-clusters (Dt1, Dt2, Dt3) through the criterion of document creation time.

Then, the filtering module 130 may perform filtering on each of the detailed clusters Dt1, Dt2, and Dt3 and specify one or several unstructured data as the learning target data.

In this way, filtering by performing clustering based on document creation time is more effective than simply subdividing the collection of unstructured data. This is because the collection of unstructured data is preferably carried out in units of relatively long cycles (eg, several months) since the amount of data must exist sufficiently to affect the learning result of the natural language processing model 120 . However, among the unstructured data collected over a relatively long period of time, not only do content created to have redundancy actually exist based on any one content, but it may also be independent content that is similar in meaning but separately created by a separate creator. . Therefore, in the latter case, even if the meaning is similar, it may be desirable to be included in the learning target data as separate information.

Therefore, after collecting unstructured data corresponding to a specific financial item in a relatively long period, a relatively short period of time based on the similarity between the unstructured data (documents) and the creation time (eg, 1 or 2 days) By performing clustering of documents within a range, the purpose of the present invention for generating related keywords according to the contents containing individual views or analyzes of multiple content creators can be achieved while eliminating redundancy.

Accordingly, the filtering module 130 may perform filtering to leave only documents corresponding to one or several document vectors in each of the detailed clusters Dt1, Dt2, and Dt3 as learning target data.

On the other hand, since the technical idea of the present invention has a feature of determining keywords related to a specific financial item through opinions, analysis, or news of a plurality of content creators on the network, changes or trends in these related keywords may affect the specific financial item or It may reflect a change in the corresponding company or a change in the perception of many people about them.

Therefore, if there is a change in keywords related to a specific financial item, it may be possible to create an investment strategy that relatively considers the change in case of recognizing it in a short time.

To this end, the item DB building module 100 may determine a related keyword for each predetermined period.

In addition, it is possible to determine related keywords with unstructured data generated within a predetermined period for each period, and store and manage the results in a predetermined DB 500 to monitor changes in related keywords of specific financial items. .

An example of this will be described with reference to FIG. 8 .

8 is a diagram for explaining changes in related keywords according to an embodiment of the present invention.

Referring to FIG. 8 , the extraction module 110 may determine a related keyword of a specific financial item (eg, S1, S2, etc.) in the above-described manner.

For example, keywords (eg, K1, K2, K3, K4) associated with the specific financial issue (eg, S1) are determined in the first period (P1), and in the second period (P2) Although the related keywords are determined again, there may be no difference from the related keywords determined in the first period (P1). That is, the related keywords determined in the second period P2 may also be the same K1, K2, K3, and K4 as those in the first period P1.

However, the related keywords determined in the third period P3 may be K1, K3, K4, and K5. That is, a new related keyword (K5) may be determined as a keyword related to the specific financial issue (eg, S1) in the third cycle, which is actually the specific financial issue (eg, S1) or a company corresponding thereto. This may mean that an issue related to the related keyword K5 has occurred, or at least users who recognized that the issue has occurred have appeared.

In this case, the extraction module 110 may perform a predetermined alarm process, and this alarm process may generate a new related keyword to the manager of the entity operating the item DB construction module 100 or to a previously designated notification target or machine. It may be a procedure to notify that it has appeared. Through this, the object receiving the notification has the effect of confirming it and quickly proceeding with the review or establishment of a new investment strategy.

9 is a flowchart illustrating a process of constructing an item-keyword relation DB by the item DB construction module 100 according to an embodiment of the present invention by way of example.

As shown in FIG. 9 , the item DB construction module 100 may perform steps S110 to S130 for each of a plurality of predefined financial investment items S (S100).

In step S110, the item DB construction module 100 may collect unstructured data corresponding to a specific financial item S.

Then, in step S120, the item DB building module 100 may determine a keyword associated with the specific financial item S based on the collected unstructured data. As described above, a context-sensitive natural language processing model may be used to determine the relevant keyword, and the learning target data used for learning the natural language processing model is not all of the collected unstructured data, but a predetermined filtering process is performed. It is as described above that it can be specified later.

In step S130, the item DB building module 100 matches the specific financial item S with keywords related to the specific financial item S and stores them in the DB 500 or items on the DB corresponding to the specific financial item S (eg, records) can be updated.

Meanwhile, the above process may be repeatedly performed when a predetermined period, that is, a predetermined period of time has elapsed.

The extraction module 110 may extract an investment impact indicator corresponding to a specific financial item, and may store the related investment impact indicator corresponding to the specific financial item in an item-indicator relation DB. An example of this will be described with reference to FIG. 10 .

10 is a diagram for explaining a process of constructing an event-indicator relation DB according to an embodiment of the present invention.

As shown in FIG. 10, steps S210 to S230 may be performed for each of a plurality of predefined financial investment items S (S200).

As described above, in step S210, the data collection module 50 can collect unstructured data corresponding to the financial item S. According to an embodiment, the data collection module 50 may collect atypical documents corresponding to keywords related to specific financial items. That is, the item DB building module 100 may determine the investment impact index using only atypical documents (referred to as first unstructured documents) corresponding to the financial item S, and may determine the financial item as well as the first unstructured documents. The investment impact index may be determined further based on unstructured documents (referred to as second unstructured documents) corresponding to the related keyword of item S. Depending on the embodiment, the investment impact index may be determined further based on unstructured documents (referred to as third unstructured documents) corresponding to the related keyword of the related keyword.

In this way, the order of related keywords is expanded, such as not only documents that mention specific financial items themselves, but also documents that mention related keywords of specific financial items, and furthermore, documents that mention keywords related to keywords related to specific financial items. Over time, documents related to specific financial issues can be expanded, and through these, investment impact indicators can be determined as described below. Documents corresponding to several related keywords (for example, if an atypical document corresponding to a related keyword of a specific financial item is defined as a secondary unstructured document or a secondary unstructured document, it corresponds to a related keyword of a related keyword of a specific financial item) Whether or not to determine the investment impact index, including up to the tertiary unstructured document (which can be defined as a tertiary unstructured document), may vary depending on the embodiment.

In addition, overlapping documents may inevitably exist between the first unstructured document (primary unstructured document) and the second unstructured document, which is the next order, and in this case, it can be treated as a fast order unstructured document.

In any case, when n (n is a natural number greater than or equal to 2) order unstructured documents including primary unstructured documents are collected by the data collection module 50, the extraction module 110 performs the financial analysis on the basis of the collected unstructured documents. A related investment impact index corresponding to item S may be determined (S220).

The extraction module 110 may use a second natural language processing model included in the natural language processing model 120 to determine an investment impact index corresponding to the financial item S based on the collected unstructured documents.

The second natural language processing model includes that each of the corresponding documents has a positive or negative effect on a specific financial item for all or some of the documents to be judged among the atypical documents collected as described above. It can be classified according to a pre-determined classification result.

The second natural language processing model may perform affect determination, that is, sentiment analysis, for each sentence included in the document as well as for the document itself. can do.

According to an example, the second natural language processing model is trained to perform sentiment analysis through a context-sensitive natural language processing model, and unstructured documents collected through the learned second natural language processing model. Sentiment analysis may be performed on all or part of them.

The sentiment analysis result output by the second natural language processing model may simply be three classes of positive, negative, or neutral, but more detailed classes (eg, strong positive, weak positive, strong negative, weak negative, neutral, etc.) can also be classified as Of course, when learning the second natural language processing model, labeling must be performed for each class to be output.

On the other hand, in the case of specifying all of the collected unstructured documents as documents to be judged to perform sentiment analysis, as described above, any one content is copied and uploaded as it is or content with only slight changes is uploaded on social media. Due to the content redundancy problem of , in practice, a single content is reproduced in multiple documents, which may cause an excessive effect on the investment impact index.

To this end, the object to which the second natural language processing model performs sentiment analysis may also be filtered by the filtering module 130, and the criteria may be the same as those described above with reference to FIG. 7 .

That is, the filtering module 130 may also perform filtering on documents to be judged, which are subjects for which the second natural language processing model performs sentiment analysis, in the same or similar manner to the filtering performed to determine the related keyword.

The filtering module 130 may perform filtering on unstructured documents corresponding to predetermined filtering conditions among the collected unstructured documents, and unstructured documents remaining after filtering may be specified as documents to be judged.

At this time, for filtering, as described above, clustering may be performed between irregular documents having a similarity between documents of a certain level or higher, and filtering may be performed to exclude a part of each of the clustered irregular document clusters from the document to be judged. .

In addition, it is possible to cluster unstructured documents in which not only the similarity but also the creation time of the unstructured document is within a predetermined time range, as described above. Also, as described above, a document vector can be used to determine the degree of similarity between unstructured documents.

In this way, the second natural language processing model may perform sentiment analysis on atypical documents after redundancy is resolved to some extent, that is, documents to be judged.

The concept of determining an investment impact index using the result of sentiment analysis on documents to be judged is shown in FIG. 11 . 11 is a diagram for explaining the concept of determining an investment impact index using atypical documents according to an embodiment of the present invention.

Referring to FIG. 11, the second natural language processing model receives each of the judgment target documents (eg, D1 to Dn) as input when the judgment target documents are specified, and the sentiment analysis result for each judgment target document (for example, For example, fD1 to fDn) may be output.

When the second natural language processing model is trained to perform output in three classes of positive, negative, and neutral, the sentiment analysis results (eg, fD1 to fDn) may also be any one of positive, negative, and neutral. there is.

Then, each of positive, negative, and neutral can be replaced with a predetermined number (eg, 1, -1, 0), and the weight for each document is defined in a predetermined way, so that investment as shown in FIG. Impact indicators can be determined. Numerical values corresponding to positive, negative, and neutral can be various embodiments.

That is, the investment impact index (F) can be defined as the product of the sentiment analysis result of each document and the weight of each document.

At this time, the weight may be determined in various ways, and according to an example, the weight may be the source of the document. For example, a weight may be determined in advance for each document source, and the weight may be determined according to the reliability of each source. For example, when a document is a source with relatively high reliability, such as a media outlet, a public institution, or a financial institution, the price may be set high, and when it is an individual user, it may be set relatively low. In addition, even when individual users are sources, the reliability of each user can be determined in various ways (eg, the number of contents produced, the number of followers, etc.), and differential weights can be assigned accordingly.

On the other hand, in the sentiment analysis for each document, as described above, when the second natural language processing model is learned, the learning data is set as the document itself, and a plurality of documents that are the learning data are labeled. If input, it may be implemented to output sentiment analysis results for the judgment target documents themselves. However, unless the documents are relatively short, it is difficult to label long documents at the time of generating training data for the documents themselves, and when positive sentences and negative sentences coexist in the document, positive sentences are collectively documented. Or it may not be easy to judge negatively.

Therefore, according to the technical concept of the present invention, the sentiment analysis result of each sentence can be used to perform the sentiment analysis result of documents, and this concept will be described with reference to FIG. 12 .

12 is a diagram for explaining the concept of performing the impact evaluation of a document through sentiment evaluation in sentence units according to an embodiment of the present invention.

Referring to FIG. 12, the second natural language processing model may perform sentiment analysis such as positive, negative, or neutral for each sentence (eg, S1 to SM) included in each of the judgment target documents Di. there is. Of course, for this, the second natural language processing model must be trained using labeled sentences as training data during learning.

Then, based on the sentiment analysis result for each sentence (eg, fs1, fs2, ..., fsM), the extraction module 110 may determine the sentiment analysis result fDi of the document Di. .

The sentiment analysis result fDi of the document Di may also be determined by multiplying the sentiment analysis result for each sentence and their weights (eg, α1 to αm), and the weights may be adaptively determined through experiments. can

In this way, when the sentiment analysis result of the document is derived through the sentiment analysis result of each sentence, there is an effect that more accurate sentiment analysis can be performed compared to learning to perform sentiment analysis on the document itself.

On the other hand, the unstructured documents collected by the data collection module 50 correspond not only to documents (first unstructured documents or secondary unstructured documents) corresponding to specific financial items, but also to keywords related to specific financial items, as described above. The range can be expanded, such as a document (second unstructured document or secondary unstructured document) or a document corresponding to a related keyword of a related keyword (third unstructured document or tertiary unstructured document).

In this case, the partial investment impact index can be determined individually for each order of documents to be judged, and the overall investment impact index can be determined based on the partial investment impact index.

An example of this will be described with reference to FIG. 13 .

13 is a diagram for explaining the concept of adaptively performing impact evaluation by reflecting the order of a document according to an embodiment of the present invention.

Referring to FIG. 13 , documents after filtering by the filtering module 130 may be D11 to D1N, D21 to D2M, and D31 to D3P.

Further, D11 to D1N may be irregular documents corresponding to a specific financial item, that is, first irregular documents, D21 to D2M may be secondary irregular documents, and D31 to D3P may be tertiary irregular documents.

In this case, the extraction module 110 individually calculates the investment impact indicators (F1, F2, F3, etc.) for each order of each unstructured document, and discriminates weights (eg, a, b) for the investment impact indicators for each order. , c) can be calculated to determine the overall investment impact index (F).

In this case, the lower the degree, the higher the weight may be set. This is because the lower the order, the content that is directly mentioned in the specific financial item, and the higher the directness, the greater the impact on the specific financial item.

Ultimately, the investment impact index can be determined by expanding the range of documents affecting the investment impact index according to the expansion of keywords while determining the investment impact index of a specific financial item through the contents produced by multiple producers on the network. In addition, by adjusting the degree of impact on the investment impact index differentially according to the degree of expansion, when users actually look at the document and determine that the relationship with a specific financial item is direct, the document has a higher impact and the relationship can be considered indirect or low. The investment impact index may be determined so as to have a low impact on the investment impact index.

Referring to FIG. 10 again, in step S230, the item DB construction module 100 matches the specific financial item S with the relevant investment impact index corresponding to the specific financial item S and stores them in the DB 500 or the specific financial item S. You can update an item (for example, a record) on the DB corresponding to S.

Meanwhile, the above process may be repeatedly performed when a predetermined period, that is, a predetermined period of time has elapsed.

Referring back to FIG. 2 , the user-indicator relationship DB building module 200 may build a user-indicator relationship DB that stores investment performance indicators of interest corresponding to each of a plurality of users.

The user-indicator relationship DB construction module 200 allows a corresponding user for each of a plurality of users who have subscribed to a predetermined service provided by using the item information providing system 10 or are pre-registered in the item information providing system 10. Interested investment impact indicators can be extracted and stored in the user-keyword relation DB.

In one embodiment, the investment impact index of interest that can correspond to each user may be a part of the investment impact index on the item-indicator relation DB. That is, the investment impact index existing on the item-indicator relation DB can be a candidate for the investment impact index of interest corresponding to a specific user.

In one embodiment, the user-indicator relationship DB building module 200 may determine an investment impact index of interest corresponding to each user through a question-and-answer procedure. For example, the user-indicator relationship DB construction module 200 may transmit a query about a preferred investment impact index to the user terminal 30, and the user terminal 30 may send an investment impact index response to the user terminal 30. It can be determined that the interest investment impact index corresponding to the user of the terminal 30 is.

In another embodiment, the user-indicator relationship DB construction module 200 may determine an investment impact index of interest for each user based on unstructured data consumed or generated by the corresponding user.

To this end, the data collection module 50 may collect unstructured data corresponding to the specific user. The unstructured data corresponding to the specific user may be unstructured data produced (eg, written) or consumed (eg, viewed) by the specific user. For example, when the specific user comments on a specific post on an internet bulletin board, both the specific post and the comment may be unstructured data corresponding to the specific user. Depending on the embodiment, the data collection module 50 collects unstructured data corresponding to a specific user, using predetermined user authentication information (e.g., For example, ID/password information, OAuth information) may be obtained from the user in advance.

The user-index relationship DB construction module 200 may extract a keyword of interest corresponding to the specific user based on the collected unstructured data corresponding to the specific user.

In an embodiment, the user-indicator relationship DB construction module 200 may extract a keyword of interest corresponding to a specific user based on the frequency of each keyword appearing in unstructured data corresponding to the specific user. For example, the user-indicator relationship DB construction module 200 may determine keywords with a frequency of appearance of more than a certain number of times or several top keywords with the highest frequency of appearance as keywords of interest. In another embodiment, the user-indicator relationship DB construction module 200 may determine a representative keyword or representative keywords of each unstructured data (document), and extract a keyword of interest based on the frequency selected as the representative keyword. may be As an example of a method of determining a representative keyword, TF-IDF or a method using the same may be cited. The user-indicator relationship DB construction module 200 may limit unstructured data used to determine a keyword of interest of a specific user based on a production time of the unstructured data (eg, upload time on a network). . For example, the user-indicator relationship DB construction module 200 may determine a keyword of interest of a specific user based only on unstructured data generated in a recent predetermined period (eg, 1 month, 3 months, etc.). In addition, while periodically performing the determination of the keyword of interest, the keyword of interest of the specific user may be periodically updated.

When a keyword of interest of a specific user is determined in the above manner, the user-indicator relationship DB construction module 200 may extract an investment item corresponding to the user's keyword of interest from the item-keyword relationship DB, and the item - An investment impact index corresponding to an investment item corresponding to the user's keyword of interest can be extracted from the index relation DB. For example, when it is determined that a keyword of interest of a specific user is keyword A, the user-indicator relationship DB construction module 200 may search and extract a related item corresponding to keyword A from the item-keyword relationship DB. Thus, if it is determined that the related items corresponding to the keyword A are item 1 and item 2, the user-indicator relationship DB construction module 200 searches the item-indicator relationship DB to determine the investment impact indicator corresponding to item 1 and item 2. Corresponding investment impact indicators can be extracted. Then, the investment impact index of interest corresponding to the specific user may be determined as an investment impact index corresponding to item 1 and an investment impact index corresponding to item 2.

Meanwhile, the update determination module 300 may observe whether the related investment impact index corresponding to each of the plurality of investment items is updated. The fact that a specific investment impact index is updated may mean that the value at the point of observation of the investment impact index is different from the previous value. The update determination module 300 receives data from a predetermined external system that provides data capable of calculating the value of the investment impact index of each financial investment item, or determines the value of the investment impact index based on data updated by itself. Calculate and compare with previous values.

In one embodiment, the update determination module 300 may periodically observe whether the associated investment impact index corresponding to each of the plurality of investment items is updated.

Alternatively, the update determination module 300 may specify an issue keyword based on unstructured data recently generated during a predetermined period, and may check whether an investment impact index corresponding to an investment item related to the specified issue keyword is updated. The issue keyword may be a keyword indicating a social issue that occurred at the time of determining an issue financial item. In addition, issue items may refer to financial items that may be directly or indirectly affected by various social issues such as politics, economy, and industry. The issue keyword may be automatically extracted from unstructured data collected within a predetermined period of time (eg, a day or a week) from the present among collected unstructured data. For example, the update determination module 300 may automatically determine the issue keyword based on various factors such as the frequency of keywords in unstructured data and the source of unstructured data. Since various prior arts (Korean Patent Application Application Nos. 10-2015-0012255, 10-2014-0081204, 10-2019-0146726, etc.) for determining these issue keywords are widely known, issue keywords based on unstructured data A description of a specific method for determining is omitted in the present specification. According to another embodiment, the update determination module 300 may receive keywords that are issues from various service-side systems such as a search service and an online product sales platform, and may specify the received keywords as issue keywords. Depending on the embodiment, when a manager or a user directly inputs a keyword for an issue determined by the manager or an issue of interest, the update determination module 300 may specify the input keyword as an issue keyword.

The update determination module 300 may determine an issue item, which is a financial investment item corresponding to the issue keyword, based on the item-keyword relation DB constructed in advance. That is, the update determination module 300 may determine an issue item corresponding to the issue keyword by searching for an item corresponding to the issue keyword in the item-keyword relationship DB.

Meanwhile, the information providing module 400 may determine a target user corresponding to the updated investment impact index from among the plurality of users based on the user-index relationship DB, and provide the updated investment impact index to the target user. It is possible to provide personalized information on investment stocks corresponding to indicators.

The information providing module 400 may search the user-indicator relationship DB to search for a user or users for whom the updated investment impact index is an investment influence index of interest, and determine the searched user as the target user. there is.

Thereafter, the information providing module 400 may provide the target user with information on investment items corresponding to the updated investment impact index. Information on a specific investment item provided to users includes the name of the item, theme, stock price flow, news related to the item, analyst analysis report related to the item, market information on the sector including the item, disclosure information, It may include various investment indices related to the stock, investment strategy including the stock, investment logic, information on the portfolio, and other various information related to the above investment stock.

As described above, according to the technical idea of the present invention, when an update occurs in a specific investment impact index, a financial item related thereto can be quickly and accurately determined. In addition, there is an effect of quickly providing information on related financial investment items to users who are determined to be interested in these investment impact indicators.

14 is a flowchart illustrating an example of a method for providing personalized item information based on artificial intelligence according to an embodiment of the present invention.

Before the process of FIG. 14 is performed, an item-indicator relationship DB for storing related investment impact indicators corresponding to each of a plurality of investment items, an item-keyword relationship DB for storing related keywords corresponding to each of the plurality of investment items, and It is assumed that a user-indicator relationship DB for storing investment impact indicators of interest corresponding to each of a plurality of users is built in advance.

Referring to FIG. 14 , the item information providing system 10 may observe whether the related investment impact index corresponding to each of a plurality of investment items is updated (S300).

Then, when a related investment index with an update is observed, a target user corresponding to the related investment index with an update is determined among a plurality of users based on the item information providing system 10 and the user-indicator relationship DB. It can (S310).

In addition, the item information providing system 10 may determine an investment item corresponding to an updated investment index based on the item-indicator relationship DB, and provide personalized information about the investment item to a target user. It can (S320).

Meanwhile, according to an implementation example, the item information providing system 10 may include a processor and a memory storing a program executed by the processor. The processor may include a single-core CPU or a multi-core CPU. The memory may include high-speed random access memory and may also include non-volatile memory such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. Access to memory by processors and other components may be controlled by a memory controller. Here, the program, when executed by a processor, may cause the item information providing system 10 according to the present embodiment to perform the above-described method.

On the other hand, the method for providing personalized event information based on artificial intelligence according to an embodiment of the present invention may be implemented in the form of computer-readable program commands and stored in a computer-readable recording medium, The control program and the target program according to the above may also be stored in a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored.

Program commands recorded on the recording medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in the software field.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, floptical disks and hardware devices specially configured to store and execute program instructions, such as magneto-optical media and ROM, RAM, flash memory, and the like. In addition, the computer-readable recording medium is distributed in computer systems connected through a network, so that computer-readable codes can be stored and executed in a distributed manner.

Examples of program instructions include high-level language codes that can be executed by a device that electronically processes information using an interpreter, for example, a computer, as well as machine language codes generated by a compiler.

The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

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

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

Claims (13)

constructing, by a computing system, an item-indicator relationship DB that stores related investment impact indicators corresponding to each of a plurality of investment items;
constructing, by the computing system, a user-indicator relationship DB for storing an investment impact index of interest corresponding to each of a plurality of users;
Observing, by the computing system, whether an associated investment impact index corresponding to each of the plurality of investment items is updated;
determining, by the computing system, a target user corresponding to the updated related investment index from among the plurality of users based on the user-index relationship DB, when a related investment index with an update is observed; and
and providing, by the computing system, personalized information about an investment item corresponding to the related investment index in which an update has occurred to the target user.
According to claim 1,
The step of constructing the user-indicator relationship DB, for each of the plurality of users,
collecting unstructured data consumed by the user; and
A method of providing item information comprising determining an investment impact index of interest corresponding to the user based on unstructured data consumed by the user.
The method of claim 1, wherein the method of providing item information,
The computing system further comprises constructing an item-keyword relationship DB for storing a related keyword corresponding to each of the plurality of investment items,
The step of constructing the user-indicator relationship DB, for each of the plurality of users,
collecting unstructured data consumed by the user;
extracting a keyword of interest of the user based on unstructured data consumed by the user;
extracting an investment item corresponding to the user's interest keyword from the item-keyword relation DB; and
An item information providing method comprising extracting an investment impact indicator corresponding to an investment item corresponding to the user's interest keyword from the item-indicator relation DB.
According to claim 1,
The step of constructing an item-indicator relationship DB including related investment impact indicators corresponding to each of a plurality of investment items,
For each of the plurality of investment items, a method for determining the related investment impact index corresponding to the investment item is performed,
The method for determining the related investment index corresponding to the investment item is,
Collecting a plurality of atypical documents corresponding to the investment item or a keyword related to the investment item; and
Determining an investment impact index corresponding to the investment item based on the collected atypical documents,
The step of determining the investment impact index corresponding to the investment item based on the collected atypical documents,
Determining the investment impact index based on the output result of the impact judgment model learned through a context-sensitive natural language processing model for all or some of the collected unstructured documents to be judged; ,
The impact judgment model is a model learned to output a classification result including whether each individual document included in the documents to be judged or each sentence included in the individual document has a positive or negative impact on investment. How to provide personal item information.
The method of claim 4, wherein the method for determining the related investment index corresponding to the investment item,
Further comprising performing filtering on unstructured documents corresponding to predetermined filtering conditions among the collected unstructured documents,
A method of providing item information, characterized in that atypical documents remaining after filtering are specified as the documents to be judged.
The method of claim 5, wherein the filtering of unstructured documents corresponding to predetermined filtering conditions among the unstructured documents collected by the system comprises:
generating a document vector for each of the collected irregular documents;
clustering atypical documents having a similarity of at least a certain level based on the generated document vectors; and
A method of providing item information, comprising performing filtering to exclude some of the clustered atypical document clusters from documents to be judged.
7. The method of claim 6, wherein the clustering of irregular documents having a similarity of at least a certain level based on the generated document vectors comprises:
A method of providing item information comprising performing clustering among unstructured documents having the degree of similarity equal to or higher than a certain level and generating time of the unstructured documents within a predetermined time range.
According to claim 4,
The step of collecting a plurality of atypical documents corresponding to the investment item or a keyword related to the investment item,
Collecting first unstructured documents corresponding to the investment item and second unstructured documents corresponding to the related keyword, respectively;
The step of determining the investment impact index corresponding to the investment item based on the collected atypical documents,
and determining based on a first investment impact index extracted based on the first unstructured documents and a second investment impact index extracted based on the second unstructured documents.
According to claim 8,
The step of determining based on the first investment impact index extracted based on the first unstructured documents and the second investment impact index extracted based on the second unstructured documents,
The method of providing stock information, characterized in that the investment impact index is determined to have different weights for each of the first investment impact index and the second investment impact index.
The method of claim 4, wherein the impact determination model,
Outputs a classification result including whether each sentence included in the individual document has a positive or negative impact on investment,
A method of providing item information, characterized in that the classification result of the individual document is determined based on the determined classification result for each sentence.
A computer program stored in a computer readable recording medium to perform the method according to any one of claims 1 to 10.
A computer readable recording medium on which a computer program for performing the method according to any one of claims 1 to 10 is recorded.
As a computing system,
processor; and
A memory for storing a computer program executed by the processor,
The computer program, when executed by the processor, causes the computing system to:
Constructing an item-indicator relationship DB that stores related investment impact indicators corresponding to each of a plurality of investment items;
Constructing a user-indicator relationship DB that stores investment impact indicators of interest corresponding to each of a plurality of users;
Observing whether the related investment impact index corresponding to each of the plurality of investment items is updated;
determining a target user corresponding to the related investment index in which an update has occurred among the plurality of users based on the user-indicator relationship DB when a related investment index in which an update has occurred is observed; and
A computing system that causes the target user to perform a method comprising providing personalized information on an investment item corresponding to the related investment index in which an update has occurred.

Images