KR102104316B1 - Apparatus for predicting stock price of company by analyzing news and operating method thereof - Google Patents

Abstract

An apparatus for predicting a stock price of a company by analyzing news according to the present invention and an operation method thereof include a word vector for the extracted word matching the preset plurality of words among the extracted at least one word and the plurality of news classification groups After calculating the similarity between news classification vectors matched to each of them, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and the newly uploaded, based on the calculated average similarity Classified news into the first news classification group among the plurality of news classification groups, and then inputting at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group Calculate the positive, neutral, and negative impact values of the newly uploaded news and then calculate Different output values can be derived according to the properties of news by determining the fluctuation rate of the stock price of a preset company based on the positive, neutral, and negative impact values.

Description

A device that predicts a company's stock price by analyzing the news and how it works {APPARATUS FOR PREDICTING STOCK PRICE OF COMPANY BY ANALYZING NEWS AND OPERATING METHOD THEREOF}

The present invention relates to an apparatus for predicting a stock price of a company by analyzing news and a method of operating the same.

The stock market has its own pricing mechanism, which is influenced by a myriad of variable data. In other words, the fundamental factors affecting the stock price are so diverse and complex, and these factors fluctuate from time to time, so there is a cycle of equations in which past fluctuation factors affect the current stock price.

Recently, due to the development of Internet technology, almost all information newly created by human beings is spread, shared, and utilized through the Internet. In addition, since information available through the Internet is digitized information, it can be easily used as input data for stock price prediction. In order to predict stock price fluctuations using so-called big data, a lot of data shared on the Internet, big data analysis using a statistical model is being performed together.

However, the prediction of stock price through conventional big data analysis could not sufficiently consider the influence between input data. Therefore, there is a need for a new kind of statistical model in order to improve the reliability of prediction results of stock price fluctuations.

According to this need, in Korean Patent Registration No. 10-1458004, an input unit that receives stock-related data as input data, allocates a plurality of inputted stock-related data to a first plurality of nodes, and provides data for each data allocated to the first plurality of nodes. A modeling unit that models an artificial neural network model by applying 1 weight to a second multiple node and applying a second weight to data allocated to the second multiple node to calculate the weighted sum and input it to the activation function to enter a single value And an output unit that outputs a high value as an output value when a single value is greater than a threshold value and outputs a low value as an output value when the single value is smaller than the threshold value.

However, the stock price prediction disclosed in Korean Patent Registration No. 10-1458004 has a problem in that, by operating one neural network for one company, the output value is presented without distinguishing the properties of news.

Therefore, it is necessary to study an algorithm that can derive different output values according to the properties of news.

An apparatus for predicting a stock price of a company by analyzing news according to the present invention and an operation method thereof include a word vector for the extracted word matching the preset plurality of words among the extracted at least one word and the plurality of news classification groups After calculating the similarity between news classification vectors matched to each of them, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and the newly uploaded, based on the calculated average similarity Classified news into the first news classification group among the plurality of news classification groups, and then inputting at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group Calculate the positive, neutral, and negative impact values of the newly uploaded news and then calculate It is intended to be able to derive different output values according to the properties of news by determining the fluctuation rate of the stock price of a preset company based on the positive, neutral, and negative impact values.

An apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention includes a news site storage unit storing a plurality of preset web site addresses where news data is uploaded, and a plurality of preset news classification groups, respectively. A vector storage unit that matches and stores a news classification vector in a match, and stores a word vector for each of a plurality of preset words, and stores a positive influence value, a neutral influence value, and a negative value of news in each of the plurality of news classification groups. A news analysis neural network storage unit that matches and stores a news analysis neural network capable of calculating an impact value, and when news is uploaded from at least one web site among the plurality of web site addresses, a number included in the newly uploaded news Extracts the sentences, and performs morpheme analysis on each of the plurality of sentences to perform A morpheme analysis unit extracting at least one word constituting each sentence, a word vector for the extracted word matching the preset plurality of words among the extracted at least one word, and each of the plurality of news classification groups After calculating the similarity between news classification vectors matched to, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and based on the calculated average similarity, the newly uploaded news A news classification unit classifying the first news classification group among the plurality of news classification groups and the input of a news analysis neural network matching the at least one word extracted from the newly uploaded news to the first news classification group By inputting the value, the positive impact value and the neutral impact value of the newly uploaded news After calculating the negative impact value it includes the calculated positive impact value, to determine the price of the stock price deungrakryul of preset companies predicted based on a neutral influence value and negative impact value portion.

In addition, an operation method of a device for predicting a stock price of a company by analyzing news according to an embodiment of the present invention comprises: maintaining a news site storage unit in which a plurality of preset website addresses to which news data are uploaded are stored; Maintaining a vector storage unit for matching and storing a news classification vector to each of a plurality of preset news classification groups, and matching and storing a word vector for each of the preset plurality of words, to each of the plurality of news classification groups Maintaining a news analysis neural network storage unit by matching and storing a news analysis neural network capable of calculating a positive impact value, a neutral impact value, and a negative impact value of news, and news from at least one website among the plurality of website addresses When is uploaded, extract a number of sentences contained in the newly uploaded news , Extracting at least one word constituting each of the plurality of sentences by performing a morpheme analysis on each of the plurality of sentences, and matching the plurality of preset words among the extracted at least one word After calculating the similarity between the word vector for the extracted word and the news classification vector matched to each of the plurality of news classification groups, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity. , Classifying the newly uploaded news into a first news classification group among the plurality of news classification groups based on the calculated average similarity, and the at least one word extracted from the newly uploaded news to the first 1 By inputting the input value of the news analysis neural network matched to the news classification group, Calculating a positive impact value, a neutral impact value, and a negative impact value of the newly uploaded news, and then determining a fluctuation rate of the stock price of a preset company based on the calculated positive impact value, the neutral impact value, and the negative impact value Includes.

An apparatus for predicting a stock price of a company by analyzing news according to the present invention and an operation method thereof include a word vector for the extracted word matching the preset plurality of words among the extracted at least one word and the plurality of news classification groups After calculating the similarity between news classification vectors matched to each of them, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and the newly uploaded, based on the calculated average similarity The classified news is classified into the first news classification group among the plurality of news classification groups, and then at least one word extracted from the newly uploaded news is input as an input value of a news analysis neural network matched to the first news classification group. Calculate the positive, neutral, and negative impact values of the newly uploaded news and then calculate Different output values can be derived according to the properties of news by determining the fluctuation rate of the stock price of a preset company based on the positive, neutral, and negative impact values.

1 is a diagram showing the configuration of an apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention.
2 is a flowchart illustrating an operation method of an apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. It should be understood that these descriptions are not intended to limit the invention to specific embodiments, and include all modifications, equivalents, or substitutes included in the spirit and scope of the invention. In describing each drawing, similar reference numerals have been used for similar components, and unless otherwise defined, all terms used in this specification, including technical or scientific terms, have common knowledge in the technical field to which the present invention pertains. It has the same meaning as commonly understood by people who have it.

1 is a diagram illustrating the configuration of a device 100 for predicting a stock price of a company by analyzing news according to an embodiment of the present invention.

As illustrated in FIG. 1, an apparatus 100 for predicting a stock price of an enterprise by analyzing news according to an embodiment of the present invention includes a news site storage unit 110, a vector storage unit 120, and a news analysis neural network storage The unit 130, the morpheme analysis unit 140, the news classification unit 150, the stock price prediction unit 160, the similar news count unit 170, the stock fluctuation rate calculation unit 180 and the news analysis neural network update unit 190 It can contain.

First, the news site storage unit 110 may store a plurality of preset web site addresses to which news data are uploaded as shown in Table 1 below.

Web site address index Website address Website address 1 www.news.google.co.kr Website address 2 www.yonhapnews.co.kr Website address 3 www.newssite.kr ... ...

Next, the vector storage unit 120 matches the news classification vector to each of a plurality of preset news classification groups and stores it as shown in Table 2 below, and matches the word vector for each of the preset plurality of words to the following table Can be saved as 3.

News classification index News classification vector News classification 1 (0.45, 0.31, -0.13, ..., 0.10) News Category 2 (-0.46, 0.13, 0.11, ..., -0.59) News Category 3 (0.05, -0.22, 0.91, ..., 0.63) ... ...

Here, for example, the news classification 1 may be “performance related news”, the news classification 2 may be “product related news”, and the news classification 3 may be “business related news”, and the news classification vector may be preset. It may consist of elements corresponding to the number of words, and each of the elements may have a real value between -1 and 1.

Word index Word vector Word 1 (0.75, 0.51, -0.17, ..., 0.75) Word 2 (0.86, -0.86, -0.46, ..., -0.35) Word 3 (0.07, 0.56, -0.60, ..., 0.64) ... ...

Here, for example, the word 1 may be 'performance', the word 2 may be 'improvement', and the word 3 may be 'deterioration', and the word vector is an element corresponding to the number of a plurality of preset words Each of the elements may have a real value between -1 and 1. The news analysis neural network storage unit 130 has a positive impact value, a neutral impact value and a news value for each of the plurality of news classification groups. A news analysis neural network capable of calculating negative influence values can be matched and stored as shown in Table 4 below.

News classification index News analysis neural network News classification 1 News analysis neural network 1 News Category 2 News Analysis Neural Network 2 News Category 3 News Analysis Neural Network 3 ... ...

Here, the news analysis neural network may be composed of N filters capable of obtaining three output values by performing N preset convolution processing on the input value. Here, N may be selected from two or more natural numbers. In addition, the convolution process includes first processing to perform convolution multiplication using a filter on the input, second processing to perform rectification linear processing on the result data of the first processing, and pooling processing on the result data of the second processing. And a third process to be performed. On the other hand, the filters used for the preset N times convolution processing may be different. When the news is uploaded from at least one web site among the plurality of web site addresses, the morpheme analysis unit 140 may use the newly uploaded news. A plurality of sentences included may be extracted, and morpheme analysis may be performed on each of the plurality of sentences to extract at least one word constituting each of the plurality of sentences. In this case, the morpheme analysis unit 140 may count the number of times of overlapping of at least one word constituting each of the plurality of sentences.

For example, in the newly uploaded news, the first sentence, “As of 11:30 am on the 15th, the company is trading at 5.91% higher than the previous trading day.” Now, the third phase of clinical trials in China has been completed by December, and it is said that an application for permission will be submitted in April of next year. ”The second sentence was extracted, and morphological analysis was performed for each sentence. The words “day”, “11:30 am”, “present”, “you company”, “pre-trade date”, “5.91%”, “rise”, “price” and “transaction” and the words “have” In the second sentence above, “Gap Company”, “China”, “Partner Company”, “Last”, “11 Days”, “Anti-Cancer Virus”, “China”, “Clinical Phase 3”, “ Coming ”,“ December ”,“ Complete ”,“ Next year ”,“ April ”,“ Permit ”,“ Application ”and“ Submit ” And the word "clarify" that predicates can be extracted. At this time, the morpheme analysis unit 140 may count the number of duplicates of the words “Gap Company” and “China” twice, and count the number of duplicates of the remaining words once.

The news classification unit 150 compares the similarity between the word vector for the extracted word matching the plurality of preset words among the extracted at least one word and the news classification vector matched to each of the plurality of news classification groups. After the calculation, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and based on the calculated average similarity, the newly uploaded news is selected from among the plurality of news classification groups. It can be classified into 1 news classification group.

For example, among the words extracted from the first and second sentences, the extracted words matching the plurality of preset words are “Gap Company”, “China”, “An anti-cancer virus”, “Clinical Phase 3 In the case of ”,“ permit ”,“ application ”and“ submit ”, the news classification unit 150 includes the above“ Gap Company ”,“ China ”,“ Anti-Virus Virus ”,“ Clinical Phase 3 ”,“ Permit ”, After calculating the similarities between the word vector matched to "application" and "submission" and the news classification vector matched to each of the plurality of news classification groups, the average value of the calculated similarities is determined as the average similarity, and the determined average The newly uploaded news may be classified into a first news classification group among the plurality of news classification groups using similarity. Here, the news classification unit 150 calculates the similarity between the word vector and the news classification vector matched to each of the plurality of news classification groups using Equation 1 below, and the average similarity is the word vector and the news classification vector. It can be calculated by adding all the similarities between each of the plurality of news classification groups and dividing by the number of similarities.

는 n개의 원소(x_i)들로 이루어진 단어 벡터 x와 n개의 원소(y_i)들로 이루어진 뉴스 분류 벡터 y 간의 유사도이다. 예컨대, 뉴스 분류부(150)는 “갑 회사”에 매칭되어 있는 단어 벡터와 상기 뉴스 분류 1에 매칭되어 있는 뉴스 분류 벡터 간의 유사도를 수학식 1을 이용하여 연산할 수 있다.From here,

It is a news similarity between the classification vector consisting of n elements y (x _i) of the word vector x with n elements (y _i) consisting of. For example, the news classifying unit 150 may calculate the similarity between the word vector matched to “A Company” and the news classification vector matched to the news class 1 using Equation (1).

The stock price predicting unit 160 inputs the at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group, thereby positively affecting the newly uploaded news, After calculating the neutral impact value and the negative impact value, it is possible to determine the fluctuation rate of the stock price of a preset company based on the calculated positive impact value, the neutral impact value, and the negative impact value. Specifically, the stock price predicting unit 160 may determine a value obtained by subtracting the calculated negative effect value from the calculated positive effect value by multiplying by 30 as a fluctuation rate of a stock price of a preset company. Here, each of the positive influence value, the neutral influence value, and the negative influence value has a real value between 0 and 1, and the total sum of the positive influence value, the neutral influence value, and the negative influence value is 1.

For example, when the calculated positive influence value is 0.3, the calculated neutral influence value is 0.5, and the calculated negative influence value is 0.2, the stock price predicting unit 160 may calculate the calculated positive influence value from 0.3. The calculated negative impact value of 0.2, which is the value obtained by subtracting 0.2, is multiplied by 30, and 3 can be determined as the fluctuation rate of the stock price of a preset company. That is, the stock price predicting unit 160 may determine 3% as the fluctuation rate of the stock price of a preset company based on the newly uploaded news.

In addition, the stock price predicting unit 160 positively affects a weight (or a weight inversely proportional to the square of the number of similar news) counted among the number of similar news counted among news searched during the preset trading day period, By applying to the neutral influence value and the negative influence value, the fluctuation rate of the stock price of the company can be determined. To this end, the similarity news counting unit 170 may count the number of similarities that have an average similarity with the newly uploaded news that is higher than or equal to a predetermined reference similarity among news searched during the preset trading day period. Here, the average similarity with the newly uploaded news is extracted from each of the word vectors for the extracted word matching the plurality of preset words among the extracted at least one word and the news searched during the preset trading day period. After calculating the similarity between word vectors for the extracted words matching the preset plurality of words among the at least one word, the average similarity for each of the news searched during the preset trading day period is calculated based on the calculated similarity. It can be obtained by calculating. Accordingly, it is possible to prevent the corporate stock price from being highly evaluated due to overlapping news.

For example, if the calculated positive impact value is 0.3, the calculated neutral impact value is 0.5, the calculated negative impact value is 0.2, and the number of similar news counted is 5, the stock price prediction unit 160 ) Is the calculated positive impact value of 0.3 multiplied by 1/5, and the calculated negative impact value of 0.2 multiplied by 1/5 is 0.6 multiplied by 30 multiplied by 30 to the rate of change in the stock price of a preset company. Can decide. That is, the stock price predicting unit 160 may determine 0.6% as the fluctuation rate of the stock price of a preset company based on the newly uploaded news.

In addition, the stock price prediction unit 160 calculates the news upload rate of the similar news for the plurality of website addresses or the website addresses to which the newly uploaded news is uploaded, and then counts the number of similar news counted. In addition, by applying a weight inversely proportional to the value multiplied by the news upload ratio to the positive influence value, the neutral influence value, and the negative influence value, the rate of change of the stock price of the company may be determined. Accordingly, it is possible to prevent the company's stock price fluctuation rate from being highly evaluated by news focused on a small number of news media.

For example, the calculated positive impact value is 0.3, the calculated neutral impact value is 0.5, the calculated negative impact value is 0.2, the number of similar news counted is 5, and the news upload rate is 1%. In the case of (that is, 0.01), the stock price predicting unit 160 calculates the negative value by multiplying the calculated positive impact value by 0.3, the number of similar news counts multiplied by the news upload rate, and 1/500. The fluctuation rate of the stock price of a preset company can be determined as 0.006 multiplied by 30, multiplied by 1/500 multiplied by 1/500. That is, the stock price prediction unit 160 may determine 0.006% as the fluctuation rate of the stock price of a preset company based on the newly uploaded news.

Thereafter, the stock price fluctuation rate calculating unit 180 may calculate the actual fluctuation rate according to the actual stock price of the company between the actual stock price of the company and a preset evaluation time point at the time the newly uploaded news is uploaded.

The news analysis neural network update unit 190, if the actual fluctuation rate for the company and the fluctuation rate determined by the stock price prediction unit 160 have different slope codes, the news analysis neural network matching the first news classification group Can be updated.

For example, if the actual fluctuation rate for the company is -3% and the fluctuation rate determined by the stock price prediction unit 160 is 2%, and has different slope codes, the news analysis neural network update unit 190 is the first The news analysis neural network matched to the news classification group may be updated.

At this time, the news analysis neural network update unit 190 has a positive effect calculated when the at least one word extracted from the newly uploaded news is input as an input value of a news analysis neural network matched to the first news classification group. By modifying the filter coefficients of the news analysis neural network matched to the first news classification group such that the value, the neutral influence value and the negative influence value have a positive influence value, a neutral influence value, and a negative influence value corresponding to the actual fluctuation rate, The news analysis neural network matched to the first news classification group may be updated.

For example, when the actual fluctuation rate for the company is -3% and the fluctuation rate determined by the stock price prediction unit 160 is 2% and has different slope codes, the news analysis neural network update unit 190 may be From the positive influence value corresponding to the actual fluctuation rate, the value obtained by subtracting the negative influence value corresponding to the actual fluctuation rate is -0.001 (ie, -0.001 multiplied by 30 is -0.03 (-3%)). Filter coefficients of the news analysis neural network matched to the first news classification group may be modified.

The apparatus 100 for predicting a stock price of a company by analyzing the news according to the present invention as described above includes a word vector and a plurality of word vectors for an extracted word matching the plurality of preset words among at least one extracted word After calculating the similarity between news classification vectors matched to each of the news classification groups, the average similarity for each of the plurality of news classification groups is calculated based on the calculated similarity, and based on the calculated average similarity, After inputting the newly uploaded news into the first news classification group among the plurality of news classification groups, at least one word extracted from the newly uploaded news is matched to the first news classification group. By inputting, after calculating the positive impact value, the neutral impact value and the negative impact value of the newly uploaded news, By determining the computed value of the positive impact Neutral impact and value to deungrakryul the share price of companies based on pre-set negative impact on the property values according to the news you can obtain different output values.

2 is a flowchart illustrating an operation method of an apparatus 100 for predicting a stock price of a company by analyzing news according to an embodiment of the present invention.

Referring to FIG. 2, in step S210, a news site storage unit storing a plurality of preset web site addresses to which news data is uploaded may be maintained.

In step S220, a vector storage unit may be maintained by matching and storing news classification vectors to each of a plurality of preset news classification groups, and matching and storing word vectors for each of the preset plurality of words.

In step S230, a news analysis neural network storage unit that matches and stores a news analysis neural network capable of calculating a positive influence value, a neutral influence value, and a negative influence value of news in each of the plurality of news classification groups may be maintained.

In step S240, when news is uploaded from at least one web site among the plurality of web site addresses, a plurality of sentences included in the newly uploaded news are extracted, and a morpheme is analyzed for each of the plurality of sentences At least one word constituting each of the plurality of sentences may be extracted by performing.

In step S250, the average similarity between the word vector matching the extracted at least one word and the news classification vector matching each of the plurality of news classification groups is calculated, and the newly uploaded news is added to the plurality of words. Among the news classification groups, the first news classification group may be classified.

In step S260, by inputting the at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group, a positive impact value and a neutral impact of the newly uploaded news After calculating the value and the negative impact value, it is possible to determine the fluctuation rate of the stock price of a preset company based on the calculated positive impact value, the neutral impact value, and the negative impact value.

An operation method of the apparatus 100 for predicting a stock price of a company by analyzing news according to an embodiment of the present invention includes at least one word extracted from the newly uploaded news among news searched during the preset trading day period. The method further includes counting the number of similar news whose average similarity with the matched word vector is equal to or greater than a preset reference similarity, and in step S260, the number of similar news counted among the news searched during the preset trading day period is counted. The inverse proportion of the stock price of the company may be determined by applying inversely proportional weights to the positive, neutral, and negative impact values.

In addition, in the step S260, after calculating the news upload ratio of the similar news for the plurality of website addresses or the website addresses to which the newly uploaded news is uploaded, the counted number of similar news is added to the number of similar news. It is possible to determine the fluctuation rate of the stock price of the company by applying a weight inversely proportional to the multiplied value of the news upload ratio to the positive influence value, the neutral influence value, and the negative influence value.

An operation method of the apparatus 100 for predicting a stock price of a company by analyzing news according to an embodiment of the present invention includes the company between the actual stock price of the company at the time when the newly uploaded news is uploaded and a preset evaluation time Calculating the actual fluctuation rate according to the actual stock price of the company, and analyzing the news matched to the first news classification group when the actual fluctuation rate for the company and the fluctuation rate determined by the stock price prediction unit have different slope signs The method may further include updating the neural network.

Also, in the updating of the news analysis neural network, a positive influence value and a neutral value calculated when the at least one word extracted from the newly uploaded news is input as an input value of a news analysis neural network matched to the first news classification group. The first news by modifying the filter coefficients of the news analysis neural network matched to the first news classification group so that the influence value and the negative influence value have a positive influence value, a neutral influence value, and a negative influence value corresponding to the actual fluctuation rate. It is possible to update the news analysis neural network matched to the classification group.

As described above, with reference to FIG. 2, an operation method of an apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention has been described. Here, the operation method of the apparatus for predicting the stock price of a company by analyzing the news according to an embodiment of the present invention may correspond to the configuration of the operation of the apparatus for predicting the stock price of the company by analyzing the news described with reference to FIG. Therefore, a more detailed description thereof will be omitted.

An operation method of an apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention may be implemented as a computer program stored in a storage medium for execution through combination with a computer.

In addition, an operation method of an apparatus for predicting a stock price of a company by analyzing news according to an embodiment of the present invention may be implemented in a form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler.

As described above, the present invention has been described by specific matters such as specific components, etc. and limited embodiments and drawings, but is provided to help the overall understanding of the present invention, and the present invention is not limited to the above embodiments , Anyone having ordinary knowledge in the field to which the present invention pertains can make various modifications and variations from these descriptions.

Therefore, the spirit of the present invention is limited to the described embodiments, and should not be determined, and all claims that are equivalent to or equivalent to the claims, as well as the claims described below, will belong to the scope of the spirit of the present invention. .

Claims (12)

A news site storage unit storing a plurality of preset web site addresses to which news data is uploaded;
A vector storage unit matching and storing news classification vectors to each of a plurality of preset news classification groups, and matching and storing word vectors for each of the preset plurality of words;
A news analysis neural network storage unit matching and storing a news analysis neural network capable of calculating a positive influence value, a neutral influence value, and a negative influence value of news in each of the plurality of news classification groups;
When news is uploaded from at least one web site among the plurality of web site addresses, a plurality of sentences included in the newly uploaded news are extracted, and a morpheme analysis is performed on each of the plurality of sentences to perform the morphological analysis. A morpheme analysis unit extracting at least one word constituting each of the sentences;
The similarity is calculated after calculating the similarity between the word vector for the extracted word matching the plurality of preset words among the extracted at least one word and the news classification vector matched to each of the plurality of news classification groups. Based on the calculated average similarity for each of the plurality of news classification groups, based on the calculated average similarity, to classify the newly uploaded news to the first news classification group of the plurality of news classification groups News classification; And
Positive and negative influence values and neutral and negative influence values of the newly uploaded news by inputting the at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group A stock price prediction unit that determines the fluctuation rate of a stock price of a preset company based on the calculated positive impact value, neutral impact value, and negative impact value after calculating
Device for predicting a company's stock price by analyzing the news containing. According to claim 1,
A similar news counting unit counting the number of similar news that has an average similarity with the newly uploaded news that is equal to or greater than a preset reference similarity among news searched during the preset trading day period;
Further comprising,
The stock price prediction unit determines a fluctuation rate of the stock price of the company by applying a weight inversely proportional to the number of similar news counted among the news searched during the preset trading day period to the positive impact value, the neutral impact value, and the negative impact value
A device that analyzes news and predicts a company's stock price. According to claim 2,
The stock price prediction unit calculates the news upload ratio of the similar news for the plurality of website addresses or the website addresses to which the newly uploaded news is uploaded, and then adds the news upload ratio to the counted number of similar news. Determine the fluctuation rate of the stock price of the company by applying a weight inversely proportional to the multiplied value to the positive influence value, the neutral influence value, and the negative influence value
A device that analyzes news and predicts a company's stock price. According to claim 3,
A stock price fluctuation rate calculating unit for calculating an actual fluctuation rate according to the actual stock price of the company between the actual stock price of the company and a preset evaluation time at the time the newly uploaded news is uploaded; And
A news analysis neural network update unit that updates the news analysis neural network matched to the first news classification group when the actual fluctuation rate for the company and the fluctuation rate determined by the stock price prediction unit have different slope codes.
Device for predicting a company's stock price by analyzing the news further comprising. According to claim 4,
The news analysis neural network update unit calculates a positive influence value, a neutral influence value, and a calculation result when the at least one word extracted from the newly uploaded news is input as an input value of a news analysis neural network matched to the first news classification group, and The first news classification group is modified by modifying the filter coefficients of the news analysis neural network matched to the first news classification group so that the negative impact value has a positive impact value, a neutral impact value, and a negative impact value corresponding to the actual fluctuation rate. Updating the matched news analytics neural network
A device that analyzes news and predicts a company's stock price. Maintaining a news site storage unit in which a plurality of preset web site addresses to which news data are uploaded are stored;
Maintaining a vector storage unit by matching and storing news classification vectors to each of a plurality of preset news classification groups, and matching and storing word vectors for each of the preset plurality of words;
Maintaining a news analysis neural network storage unit matching and storing a news analysis neural network capable of calculating a positive influence value, a neutral influence value, and a negative influence value of news in each of the plurality of news classification groups;
When news is uploaded from at least one web site among the plurality of web site addresses, a plurality of sentences included in the newly uploaded news are extracted, and a morpheme analysis is performed on each of the plurality of sentences to perform the morphological analysis. Extracting at least one word constituting each of the sentences;
The similarity is calculated after calculating the similarity between the word vector for the extracted word matching the plurality of preset words among the extracted at least one word and the news classification vector matched to each of the plurality of news classification groups. Based on the calculated average similarity for each of the plurality of news classification groups, based on the calculated average similarity, to classify the newly uploaded news to the first news classification group of the plurality of news classification groups step; And
Positive and negative influence values and neutral and negative influence values of the newly uploaded news by inputting the at least one word extracted from the newly uploaded news as an input value of a news analysis neural network matched to the first news classification group After calculating the step of determining the fluctuation rate of the stock price of a preset company based on the calculated positive impact value, neutral impact value and negative impact value
Method of operating the device to predict the stock price of the company by analyzing the news containing. The method of claim 6,
Counting the number of similar news among the news searched during the preset trading day period, the average similarity with the newly uploaded news being equal to or greater than a preset reference similarity;
Further comprising,
The step of determining the stock price fluctuation rate applies a weight that is inversely proportional to the number of similar news counted among news searched during the preset trading day period to the positive impact value, the neutral impact value, and the negative impact value, so that the company's stock price fluctuation rate To decide
How the device works to predict the company's stock price by analyzing the news. The method of claim 7,
The step of determining the stock price fluctuation rate calculates the news upload ratio of the similar news for the plurality of website addresses or the website addresses to which the newly uploaded news is uploaded, and then adds the news to the counted number of similar news. Determine the fluctuation rate of the stock price of the company by applying weights inversely proportional to the value multiplied by the upload rate to the positive, neutral, and negative impact values.
How the device works to predict the company's stock price by analyzing the news. The method of claim 8,
Calculating an actual fluctuation rate according to the actual stock price of the company between the actual stock price of the company and a preset evaluation time point when the newly uploaded news is uploaded; And
Updating the news analysis neural network matched to the first news classification group if the actual fluctuation rate for the company and the fluctuation rate determined by the stock price prediction unit have different slope codes.
A method of operating a device for predicting a company's stock price by analyzing the news further including the. The method of claim 9,
In the updating of the news analysis neural network, a positive influence value and a neutral influence value are calculated when the at least one word extracted from the newly uploaded news is input as an input value of a news analysis neural network matched to the first news classification group. And the first news classification group by modifying the filter coefficients of the news analysis neural network matched to the first news classification group so that the negative impact value has a positive impact value, a neutral impact value, and a negative impact value corresponding to the actual fluctuation rate. To update the news analysis neural network that is matched on
How the device works to predict the company's stock price by analyzing the news. A computer-readable recording medium recording a program that causes a computer to perform the method of any one of claims 6 to 10. delete

Images