KR20220101817A - Method and system of an artificial intelligence for predicting financial information - Google Patents

Abstract

The present invention provides an artificial intelligence system for predicting financial information, comprising: a financial data collection unit that collects data to be analyzed by extracting finance-related meta information and character information by linking with at least one of a plurality of Internet media, media providing financial services, and news web pages of media companies through a communication network; a morpheme analysis unit that analyzes morphemes by dividing the financial-related character information into lexical meaning and grammatical meaning; a syntax analysis unit that constructs a parsing tree through syntax analysis for the data to be analyzed that has undergone the morpheme analysis; a sentence analysis unit that generates meaning for each word in the parsing tree and then assigns the meaning; a financial information extraction unit that extracts financial information based on the result of the sentence analysis unit; and a financial prediction information generation unit that generates financial prediction information through comparison between the extracted financial information and reference data managed by preset classification depending on fluctuations in stock prices and financial indices. Accordingly, the prediction accuracy of financial information can be increased.

Description

An artificial intelligence system for predicting financial information and a method for predicting financial information using the same

The present invention relates to an artificial intelligence system capable of predicting financial information based on natural language processing and a financial information prediction method using the same.

Recently, due to the rapid development of IT technology, the spread of terminals such as smart phones is expanding. Accordingly, the user can process various financial services online without directly visiting a bank or the like using the terminal. In addition, the number of financial transactions using electronic payment methods such as credit cards or simple payment services is increasing compared to existing financial transactions through cash.

Due to such changes in the financial transaction environment, each financial institution is developing and promoting various types of financial products. However, as the number of various payment methods and financial products increases, the related information also increases exponentially, so it is difficult for users to select a suitable financial product.

Meanwhile, information related to the financial industry is divided into numerical data such as financial indices and interest rates, and text data such as economic analysis reports and industry-related news.

Of these, text data is regarded as a unique area where the intuition of securities analysts is important, unlike numerical data that econometrics and mathematicians have actively researched.

However, due to the discrepancy between the text data and the mathematical model, the probability of making a wrong decision increases in the financial industry, which requires quick judgment, such as canceling derivatives. This can lead to huge losses for investors or companies.

Recently, with the development of analysis technologies such as hardware and machine learning, it is possible to find non-reflected information in text data and incorporate it into a model in a quantitative form. Based on this, information analysis technology in various fields is being developed.

Republic of Korea Patent Publication No. 10-2019-0082151 (published on July 9, 2019.)

The present invention provides an artificial intelligence system for predicting financial information that can collect data composed of various types of characters related to finance and extract and predict financial information based on the collected data, and a financial information prediction method using the same.

In addition, the present invention provides an artificial intelligence system capable of predicting financial information by comparing extracted financial information, stock price, and classification data classified according to the fluctuation range according to the fluctuation range, and a financial information prediction method using the same.

As a technical means for achieving the above technical problem, the artificial intelligence system for predicting financial information according to an embodiment of the present invention includes at least one or more of a plurality of Internet media, media providing financial-related services, and news web pages of media companies; A financial data collection unit that collects analysis target data by extracting financial-related meta information and text information by interworking through a communication network, a morpheme analysis unit that analyzes morphemes by dividing the financial-related text information into lexical and grammatical meanings; A syntax analysis unit for constructing a parsing tree through syntactic analysis of the analysis target data that has undergone the morpheme analysis; A financial information extraction unit that extracts financial information based on It may include a generator.

According to an embodiment of the present invention, the syntax analysis unit may construct a parsing tree by performing syntax analysis through a neural network-dependent syntax analyzer and a recursive algorithm suitable for a Korean language structure.

According to an embodiment of the present invention, the sentence analysis unit analyzes the meaning from the syntax analysis result of the syntax analysis unit using a distribution hypothesis and a vector space model, and then interprets the sentence based on the analysis result, and explains the sentence Based on this, a vector sequence in which a sentence is embedded can be generated.

According to an embodiment of the present invention, the financial information extraction unit may calculate a context vector corresponding to financial information by performing pretraining with an attention-based transformer using the embedded vector sequence.

As a technical means for achieving the above technical problem, a financial information prediction method using an artificial intelligence system in an embodiment of the present invention is a financial information prediction device, a plurality of Internet media, media and news websites providing financial-related services Collecting analysis target data by extracting financial-related meta information and text information by interworking with at least one of the pages through a communication network, and analyzing morphemes by dividing the financial-related text information into lexical and grammatical meanings; , constructing a parsing tree through syntactic analysis of the analysis target data that has undergone the morpheme analysis, generating a meaning for each word in the parsing tree and then giving it, and based on the result of the sentence interpretation unit It may include extracting financial information, and generating financial prediction information through comparison between the extracted financial information and financial information managed in a preset classification according to the fluctuation range of stock prices and financial indexes.

According to an embodiment of the present invention, the giving step analyzes the meaning in the parsing tree using a distribution hypothesis and a vector space model, and then analyzes the sentence based on the analysis result of the meaning, It is possible to create a vector sequence in which a sentence is embedded based on the explanation.

According to an embodiment of the present invention, in the step of extracting the financial information, a context vector corresponding to the financial information is calculated by performing pretraining with an attention-based transformer using the embedded vector sequence, , the generating of the financial prediction information may include determining which category the context vector belongs to by applying the context vector to the following equation, and generating financial prediction information based on the determined result.

According to the above-described embodiment of the present invention, financial prediction information can be generated by reflecting the text information by collecting data composed of various types of finance-related text and extracting and predicting financial information based on the collected data. Therefore, it is possible to increase the prediction accuracy of financial information.

1 is a diagram illustrating an artificial intelligence system for predicting financial information according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining the configuration of a parsing tree output from a classification analysis unit of an artificial intelligence system according to an embodiment of the present invention.
3 is a flowchart illustrating a financial information prediction process in an artificial intelligence system according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The following detailed description is provided to provide a comprehensive understanding of the methods, apparatus, and/or systems described herein. However, this is merely an example and the present invention is not limited thereto.

In describing the embodiments of the present invention, if it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. And, the terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to intentions or customs of users and operators. Therefore, the definition should be made based on the content throughout this specification. The terminology used in the detailed description is for the purpose of describing embodiments of the present invention only, and should in no way be limiting. Unless explicitly used otherwise, expressions in the singular include the meaning of the plural. In this description, expressions such as “comprising” or “comprising” are intended to indicate certain features, numbers, steps, acts, elements, some or a combination thereof, one or more other than those described. It should not be construed to exclude the presence or possibility of other features, numbers, steps, acts, elements, or any part or combination thereof.

Hereinafter, an artificial intelligence system for predicting financial information and a method of predicting financial information using the same according to an embodiment of the present invention will be described with reference to the accompanying drawings.

1 is a diagram illustrating an artificial intelligence system for predicting financial information according to an embodiment of the present invention, and FIG. 2 illustrates the configuration of a parsing tree output from a division analysis unit of an artificial intelligence system according to an embodiment of the present invention It is a drawing for

Prior to the description according to an embodiment of the present invention, the artificial intelligence system for predicting financial information may operate based on natural language processing, which implements a natural language interface that can communicate with various computing devices in human language. It means a technology to do this, and it may mean that the process of understanding symbols and calculating symbols occurring in a human brain organ is simulated by a computing device.

As shown in Figure 1, the artificial intelligence system for predicting financial information can be implemented as a natural language processing method made through a financial information algorithm to which a natural language processing algorithm is applied using deep learning to which an optimization mathematical theory is applied. .

The apparatus 100 for predicting financial information of an artificial intelligence system for predicting financial information according to an embodiment of the present invention may operate based on a natural language processing method implemented using Python.

The financial information prediction apparatus 100 may calculate a vector sequence in which a sentence is embedded by applying a Skipgram-based Bidirectional Embedding Representation Transformer (BERT) in PYTORCH or TENSORFLOW. In this case, the necessary data may use the language information database 150 constructed using Korean Wikimedia, Netflix, or collected newspaper article information.

To this end, the financial information prediction apparatus 100 according to an embodiment of the present invention collects text information through crawling, analyzes the morpheme and syntax on the collected text information, interprets the sentence, and then provides financial information based on this. It can be extracted to generate financial forecast information.

To this end, the financial information prediction apparatus 100 includes a financial data collection unit 102 , a morpheme analysis unit 104 , a syntax analysis unit 106 , a sentence analysis unit 108 , a financial information extraction unit 110 , and a financial prediction unit. It may include an information generating unit 112 and the like.

First, the financial data collection unit 102 may collect financial data for natural language processing of text data.

In an embodiment of the present invention, the financial data collection unit 102 may collect financial data through a crawling procedure in order to collect a large amount of valuable atypical data. Specifically, the financial data collection unit 102 includes a finance-related crawler 102a, and through crawling using the finance-related crawler 102a, financial-related meta information and text information from various Internet media and news webpages It is possible to collect data to be analyzed by extracting it.

The financial data collection unit 102 according to an embodiment of the present invention provides various Internet media, for example, financial information service, exchange rate change service, stock price related service, interest rate related service, etc. through a communication network. It is possible to collect analysis target data by extracting financial-related meta information and text information in conjunction with media.

The morpheme analyzer 104 may divide a symbol between blanks of character information in the analysis target data into a stem and a stem. Specifically, the morpheme analyzer 104 may analyze the morpheme by dividing the character information collected through crawling into minimum semantic units, that is, lexical meaning and grammatical meaning.

In particular, in the embodiment of the present invention, the morpheme analyzer 104 performs morpheme analysis on the data to be analyzed in consideration of part-of-speech using a morpheme analyzer module, for example, a morpheme analyzer module such as Hannanum, Kkma, Komoran, Mecab, and Twitter. can be carried out.

The syntax analysis unit 106 may construct a parsing tree as shown in FIG. 2 by performing syntax analysis on the analysis target data that has undergone morphological analysis.

In addition, the syntax analysis unit 106 may analyze the words constituting the sentence as components or semantic domains as a process of analyzing the component meaning or grammatical meaning of an adjective or a verb.

In the embodiment of the present invention, the syntax analysis unit 106 is a deep neural network (RNN), such as a Recursive Neural Network (RNN), a Syntactically United Recursive Neural Network (SU-RNN), a Convolution Neural Network (CNN), or a Long/Short Term Memory (LSTM). Syntax analysis may be performed on the analysis target data using a syntax analysis program (not shown) using a learning technique.

In addition, in an embodiment of the present invention, the syntax analysis unit 106 applies a Neural Network Dependency Parser suitable for the Korean language structure, and performs syntax analysis through a method using a recursive algorithm. can do. Here, the recursive algorithm may refer to a recursive calling method in which the own function is called again within the function.

The neural network-dependent syntax analyzer is a one-hot method in which words and features extracted from the current state of the data to be analyzed are expressed as 1 and the rest are expressed as 0 only for the corresponding word or feature. -hot) in the form of an input layer, and a projection layer that reduces the dimension by pre-training the word expression of a neural network language model (NNLM) with respect to the input layer, This may be performed using a neural network of a hidden layer that performs non-linear activation on the projection layer and an output layer that calculates a probability for each state.

The sentence interpretation unit 108 may generate a meaning for the structures generated by the syntax analysis, and then give it.

That is, the sentence interpretation unit 108 may generate an accurate structure in a manner in which the meaning of each word is combined with each other by matching an appropriate entity to each word output through syntax analysis. At this time, the sentence interpretation unit 108 may select an appropriate entity for each word using a knowledge database (not shown) and then create an accurate structure by corresponding the entity to each word.

In an embodiment of the present invention, the sentence interpretation unit 108 uses a distributional hypothesis and a vector space model to analyze the meaning in the syntax analysis result.

In this case, the distribution hypothesis is a hypothesis that words appearing in a similar context have similar meanings, and the vector space model expresses each word or phrase belonging to a document set as a vector in a vector space to define the distance between the vectors. have.

Through this, the sentence interpretation unit 108 generates a meaning through the distribution hypothesis that words appearing in similar contexts have similar meanings through the distribution hypothesis, and calculates the distance between words or phrases of the vector space model. By quantifying the meaning of each, it is possible to generate a meaning for the parsing tree structures generated by the syntax analysis and then assign it.

In particular, the sentence interpretation unit 108 according to an embodiment of the present invention analyzes the sentence using a financial specialized sentence interpreter (not shown) based on deep learning such as a Dynamic Memory Network (DMN) and a Bayesian Classifier. After implementation, based on this, as shown in FIG. 2 , a vector sequence in which a sentence is embedded can be generated.

The financial information extraction unit 110 according to an embodiment of the present invention performs pretraining with an attention-based transformer or autoencoder to which an embedded vector sequence is input, and performs pretraining to a context vector (Context vector) ) can be calculated.

In this case, the context vector includes information on the entire article and sentence, and can be used for deep learning applications, and can be an input vector for reinforcement learning that views changes in stock prices or financial indicators as a result.

In an embodiment of the present invention, financial information may be managed in a hexadecimal classification according to the fluctuation range of stock prices and financial indices, and the modeling model applied in the embodiment of the present invention is financial information and financial information managed in a hexadecimal classification. A context vector input from the extractor 110 may be used.

The financial prediction information generating unit 112 may generate financial prediction information through fine tuning based on the financial information managed by hexadecimal classification and the context vector obtained through pre-processing. Specifically, the financial prediction information generating unit 112 performs mathematical modeling on the financial information managed by hexadecimal classification and the context vector obtained through pre-processing using Equation 1 below to determine which of the hexadecimal classifications the context vector is. By determining whether it belongs to a classification, financial information can be predicted based on natural language processing.

[Equation 1]

는 i그룹의 기댓값 (

)이며,

는 데이터(즉, 추출한 콘텍스트 벡터)가 i 그룹에 속할 확률이며,

는 예측치가 i 그룹에 속한다는 가설이 진실일 확률이다. In Equation 1 above

is the expected value of group i (

) and

is the probability that the data (that is, the extracted context vector) belongs to group i,

is the probability that the hypothesis that the predicted value belongs to group i is true.

는 6진 분류로 나누어진 금융 정보 그룹의 기댓값일 수 있다.in other words,

may be the expected value of the financial information group divided into hexadecimal classification.

Through Equation 1 as described above, the financial prediction information generation unit 112 determines to which group of reference data the context vector related to the financial information extracted by the financial information extraction unit 110 belongs to, and then, based on the determination result, information can be predicted.

In an embodiment of the present invention, the reference data for financial information divided into hexadecimal classification is for the fluctuation range of stock prices and financial indices, and for example, in the case of stock prices, stock price rise, fall, rise rate, decline rate, etc. However, the present invention is not limited thereto.

That is, if the context vector corresponding to the financial information extracted by the financial information extraction unit 110 based on this information is related to the stock, the financial prediction information generation unit 112 divides the context vector into hexadecimal classification through Equation 1 It may be determined which part of the reference data for financial information prediction belongs, and financial prediction information may be generated based on the determination result.

A process of predicting financial information by the financial information prediction apparatus 100 according to an embodiment of the present invention as described above will be described with reference to FIG. 3 .

3 is a flowchart illustrating a financial information prediction process in an artificial intelligence system according to an embodiment of the present invention.

As shown in FIG. 3 , first, the financial information prediction apparatus 100 collects analysis target data, that is, financial-related meta information and financial-related text information, extracts financial information through a natural language processing process, and models it on the extracted financial information. The model may be applied to generate financial forecasting information, for example, financial forecasting information such as indices, interest rates, and the like.

To this end, the financial data collection unit 102 of the financial information prediction apparatus 100 according to an embodiment of the present invention establishes a communication network with at least one of a plurality of Internet media, media providing financial-related services, and news web pages of media companies. Through interlocking, financial-related meta information and financial-related text information are extracted to collect analysis target data (S300).

Then, the financial information prediction apparatus 100 analyzes the morphemes by dividing the character-related data of the analysis target data into lexical meaning and grammatical meaning through the morpheme analysis unit 104 ( S302 ).

Then, the financial information prediction apparatus 100 constructs a parsing tree by parsing the analysis target data that has undergone morphological analysis through the syntax analysis unit 106 ( S304 ). In this case, the financial information prediction apparatus 100 may perform syntax analysis through a neural network-dependent syntax analyzer and a recursive algorithm suitable for the Korean language structure.

Thereafter, the financial information prediction apparatus 100 generates a meaning for each word in the parsing tree through the sentence interpretation unit 108 and gives it to each word (S306). At this time, the financial information prediction apparatus 100 analyzes the meaning from the syntax analysis result using the distribution hypothesis and the vector space model, and interprets the sentence based on the analyzed meaning, and then embeds the sentence based on the explanation of the sentence. One vector sequence can be created.

Then, the financial information prediction apparatus 100 extracts financial information using the embedded vector sequence output from the sentence interpretation unit 108 through the financial information extraction unit 110 (S308). That is, the financial information prediction apparatus 100 may calculate a context vector (weight vector) corresponding to financial information by performing pretraining with an attention-based transformer using the embedded vector sequence.

Thereafter, the financial information prediction apparatus 100 generates financial prediction information by applying the context vector to the modeling model as in Equation 1 through the financial information prediction information generating unit 112 ( S310 ).

Meanwhile, combinations of each block in the accompanying block diagram and each step in the flowchart may be performed by computer program instructions. These computer program instructions may be embodied in a processor of a general purpose computer, special purpose computer, or other programmable data processing equipment, such that the instructions performed by the processor of the computer or other programmable data processing equipment are described in each block of the block diagram. It creates a means to perform functions.

These computer program instructions may also be stored in a computer-usable or computer-readable recording medium (or memory), etc., which may direct a computer or other programmable data processing equipment to implement a function in a specific manner, so that the computer is available. Alternatively, the instructions stored in the computer-readable recording medium (or memory) may produce an article of manufacture including instruction means for performing the functions described in each block of the block diagram.

And, since the computer program instructions may be mounted on a computer or other programmable data processing equipment, a series of operating steps are performed on the computer or other programmable data processing equipment to create a computer-executed process to create a computer or other program. It is also possible that instructions for performing the possible data processing equipment provide steps for performing the functions described in each block of the block diagram.

In addition, each block may represent a module, segment, or part of code including at least one or more executable instructions for executing specified logical function(s). It should also be noted that, in some alternative embodiments, it is also possible for the functions mentioned in the blocks to occur out of order. For example, it is possible that two blocks shown in succession are actually performed substantially simultaneously, or that the blocks are sometimes performed in the reverse order according to the corresponding function.

100: financial information prediction device
102: financial data collection unit
104: morpheme analysis unit
106: parsing unit
108: sentence interpretation unit
110: financial information extraction unit
112: financial forecast information generation unit

Claims (6)

A financial data collection unit that collects data to be analyzed by extracting financial-related meta information and text information by linking with at least one of a plurality of Internet media, media providing financial-related services, and news web pages of media companies through a communication network;
a morpheme analysis unit that analyzes morphemes by dividing the financial-related text information into lexical and grammatical meanings;
a parsing unit for constructing a parsing tree by parsing the analysis target data that has undergone the morpheme analysis;
a sentence interpretation unit that generates a meaning for each word in the parsing tree and gives it;
a financial information extraction unit for extracting financial information based on the result of the sentence interpretation unit;
An artificial intelligence system for predicting financial information, comprising a financial prediction information generation unit that generates financial prediction information through comparison between the extracted financial information and reference data managed by a preset classification according to the fluctuation range of the stock price and financial index.
According to claim 1,
The sentence interpretation unit,
Finance that analyzes the meaning of the syntax analysis result of the syntax analysis unit using a distribution hypothesis and a vector space model, analyzes the sentence based on this, and generates a vector sequence in which the sentence is embedded based on the explanation of the sentence Artificial intelligence system for information prediction.
3. The method of claim 2,
The financial information extraction unit,
An artificial intelligence system for predicting financial information that calculates a context vector corresponding to financial information by performing pretraining with an attention-based transformer using the embedded vector sequence.
In the financial information prediction device, collecting data to be analyzed by extracting financial-related meta-information and text information by interworking with at least one of a plurality of internet media, media providing financial-related services, and news web pages of media companies through a communication network Wow,
analyzing the morphemes by dividing the financial-related text information into lexical and grammatical meanings;
constructing a parsing tree by parsing the analysis target data that has undergone the morpheme analysis;
creating a meaning for each word in the parsing tree and giving it;
extracting financial information based on the result of the sentence interpretation unit;
The method of predicting financial information using an artificial intelligence system, comprising the step of generating financial prediction information by comparing the extracted financial information with the financial information managed in a preset classification according to the fluctuation range of stock prices and financial indexes.
5. The method of claim 4,
The giving step is
After analyzing the meaning in the parsing tree using a distribution hypothesis and a vector space model, analysis of the sentence is performed based on the analysis result of the meaning, and a vector sequence in which the sentence is embedded based on the explanation of the sentence is generated. A method of predicting financial information using an artificial intelligence system.
6. The method of claim 5,
The step of extracting the financial information,
Using the embedded vector sequence, an attention-based transformer and pretraining are performed to calculate a context vector corresponding to financial information,
The step of generating the financial forecast information includes,
A method for predicting financial information using an artificial intelligence system for determining which category the context vector belongs to by applying the context vector to the following equation, and generating financial prediction information based on the determined result.

Images