KR100853022B1 - Article automatic generation method and device - Google Patents

Abstract

 The present invention is a more accurate and significantly faster real-time concept than the existing editing method for urgent press releases by automating the whole process while attaching abundant additional data and useful analytical opinions to help information users in creating and reporting articles related to company disclosure. The article relates to a method for automatically generating articles and apparatus thereof.

In the present invention, search and classify data provided from information sources connected through a communication network. Generate and store meta data about the data. The content of the data is parsed to determine whether missing data is present and recorded in the corresponding item of the metadata. If there is no missing data as a result of the determination, the parsing content of the data is stored. Processing data is generated based on the data. Numerical data included in the processed data is converted into sentence data.

News, Disclosure Information, Stocks, Real Time

Description

METHOD AND APPARATUS FOR AUTOMATICALLY GENERATING ARTICLES}

1 illustrates an example of a conventionally provided article.

2 is a block diagram of a schematic automatic article providing system according to an embodiment of the present invention.

3 is a schematic block diagram of an automatic article generating apparatus according to an embodiment of the present invention.

4 is a flowchart illustrating a method for automatically generating articles according to an embodiment of the present invention.

5 to 10 show an example of data generated and referenced in the article automatic generation process according to the present invention in the case of performance disclosure.

11 is an example showing an article created according to the present invention in the case of order disclosure.

<Explanation of symbols for the main parts of the drawings>

100: article generator

200: information provider

300: communication network

400: customer terminal

500: administrator terminal

The present invention relates to a method and apparatus for automatically generating and providing an article, and more particularly, in the generation and reporting of an article related to a company disclosure, while additional information and analytical opinions are added to help an understanding of information users. The present invention relates to a method for automatically generating articles and a device for providing articles quickly.

Press articles of online newspapers that provide articles related to securities have a large proportion of press articles (straight articles) that should be urgently reported even though they include commentary articles or interview articles of existing offline newspapers. As shown in Figure 1 below, most of the manual work by the hand of the reporter has been a lot of time, labor and cost has been taken.

Examples of the straight article include articles related to the company's performance, order related articles, equity disclosure related articles, patent disclosure related articles, and other articles stating important management matters of the company.

Figure 1 shows the performance-related disclosure information and the article produced based on this in order to illustrate the above example by way of example.

Referring to FIG. 1, the performance-related disclosure information 10 of Handok Pharmaceuticals (stock unique number: 002390) was provided on October 23, 2006 at 15:26:57. Existing online newspapers have prepared a press article based on the performance-related disclosure information (10) of Handok Pharmaceuticals (stock unique number: 002390) and provided it to the customer terminal.The reporting time is a short article citing the disclosure information. After at least 3 to 55 minutes, you can see that you are reporting.

That is, in the case of A newspaper 12, the reporting time is 15:29:59 seconds on October 23, 2006, and in the case of B newspaper 14, the reporting time is 15:35:18 seconds on October 23, 2006 In the case of (16), the reporting time was 16:22:17 on October 23, 2006, and it took 3 minutes, 2 seconds, 8 minutes, 21 seconds, and 55 minutes and 30 seconds, respectively. There has been a problem that only those who reported the original disclosure once more cannot use the information to increase satisfaction with the article.

Even if newspapers automate some of the processes of creating and delivering editors, there is a limitation that time is delayed if manual work is involved in the rest, thereby degrading the value as an article in the creation of a securities article that struggles for minutes and seconds.

On the other hand, as a prior art, there is patent No. 0416953, "System News Providing System and Method thereof (hereinafter referred to as Prior Patent)". The preceding patent has a structure for generating news corresponding to logic preset by the 'user or system operator' with respect to the investment information.

However, in the case of the prior patent, since the structure of generating a fixed phrase for the event corresponding to the predetermined logic set by the user or the system operator to always flow the information flowing out, it is complete and quantitatively and qualitatively enough to be called an article. Falls. In addition, if you use the structure that only the user receives the logic set by the user, it is no different from the real-time search function used in the current home trading system (HTS) rather than news. And the result it shows is fragmentary, and it is hard to say that it is an article that is meaningful only to some users who are trading stocks, such as the upper and lower limit information, and the selling amount information. In addition, since the base data, which is the basis of news generation, is also based on a single data such as quote information, the generated article may be regarded as a simple alarm device rather than a news.

In addition, the preceding patent determines whether news is generated or not, corresponding to logic in a single data, which means that only 'information that a system operator or user thinks to be meaningful' is produced as an article. This is to leave 'articles' at the discretion of the system operator or the user, and to miss a lot of important articles, and to generate different news for each client, which causes information asymmetry. There is this. In addition, the preceding patent asks 'is there a case corresponding to logic?' Is an important determinant and content of news creation, so the content to be delivered is the occurrence of logic-related events.

Moreover, the preceding patent has the advantage that it does not cause a load on the network by encoding the system news to 6 digits and delivering it to the client, but, paradoxically, the number of items should be separated within 3 digits and the logic is 3 There is a fatal shortcoming that it cannot handle more than 1000 sports or more than 1000 different logic because they must be distinguished within the seat. In addition, this structure has a limitation in that it does not take into account the load of downloading items or logic changes every time the client connects to the server.

In addition, even securities information companies that currently use automated systems could not process press releases to produce press releases such as newspaper articles. First, disclosure information, unlike simple quotes or quotes processing information, is characterized by the complexity and variety of its contents. Due to the difficulty of automating or configuring and managing the server, and secondly, there are errors in description in the contents disclosed by companies, but it is not easy for an information provider to find errors mixed in a large amount of data. Others, such as analytical opinions and processing information, require considerable skills such as complex logic management, sentence semantic parsing, and synthesis, but the language processing technology for commercial services is not generalized.

The present invention has been made in view of the above circumstances, and its purpose is to automate the whole process by attaching abundant additional data and analytical opinions to help the users of information in the creation of news related to company disclosure, rather than the conventional manual editing method. The present invention provides an automatic article automatic generation method and apparatus for providing an accurate and extremely fast real-time article.

Another object of the present invention is to provide an analysis article that can be used as a basis for rich and comprehensive opinions or investment decisions by showing all the data to be searched by the user through linkage of base data, past cumulative data, and processed data. To provide an article for automatic generation method and apparatus therefor.

It is another object of the present invention to exclude any arbitrariness by generating articles for all disclosures unless there is an error, and articles processed in the device are published to all clients equally so that information asymmetry does not occur. The present invention provides a method and an apparatus.

Another object of the present invention is to provide a method and apparatus for automatically generating articles that can provide abundant contents and adhere to the principle of reporting without missing the essential elements of the 6th principle.

Another object of the present invention is to process all the processes in a batch does not cause unnecessary load on the client, and the network load problem is solved at the level of a normal web page, so a normal web server can be serviced without difficulty. In the unlikely event that a large number of users are connected at the same time, a separate news provider or a separate server to solve the problem to provide a method for automatic article generation and its device.

Article automatic generation method according to a first aspect of the present invention for achieving the above object,

a) retrieving, collecting and classifying data provided from information sources connected through a communication network;

b) generating and storing metadata about the data;

c) after parsing the contents of the data, determining whether missing data exists and recording the corresponding item in the metadata;

d) if there is no missing data as a result of the determination of step c), storing the parsing content of the data;

e) generating processing data based on the data; And

f) converting numerical data contained in the processed data into sentence data.

In this case, in the step d), when the data is accumulated and stored as a result of parsing the data, it is preferable to record a predetermined status indication on whether to generate the accumulated data of the metadata.

In addition, in step d), it is preferable to check whether there is an error in description of the parsing content and to perform step e) when there is no error in the description.

Furthermore, when the data is quarterly earnings data in step e), the method may further include generating processing data in association with past quarterly earnings.

The step f) may include extracting a spoken word from the numerical data, selecting a spoken word according to the increase and decrease of the numerical data, and determining analysis information according to a combination of the selected spoken word and the spoken word; And completing a sentence by inserting a search and a mother according to a preset rule between the idiom and the verb.

The processing data may include at least one of a stock price return ratio, an operating profit margin, and a equity return.

Article automatic generation device according to a second aspect of the present invention for achieving the above object,

 A search and classification module for searching, collecting, and classifying data generated from an information provider providing at least one disclosure information connected to a communication network based on a given index word;

A parsing and validation module for parsing desired content from the retrieved data and validating the parsed data;

A data processing unit including a processing module for generating processing data based on the data on which the validity check has been performed; And

And an article generation unit for converting numerical data included in the processing data provided from the data processing unit into sentence data.

The data processing unit may further include a cumulative data DB for accumulating and storing a part of the data and a processing data DB for storing processing data processed by the processing module.

The article generator may further include a tabulation module that generates at least one of a table and a graph based on data of the cumulative data DB and the processed data DB.

The article generating unit extracts a statement from the numerical data, selects a word according to the increase or decrease of the numerical data, determines analysis information according to the combination of the selected word and the word, and then follows the preset rule between the word and the word. The numerical data is converted into sentence data by inserting a search and a mother.

(Example)

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 is a block diagram of a schematic automatic article providing system according to an embodiment of the present invention, Figure 3 is a schematic block diagram of the automatic article generation apparatus according to an embodiment of the present invention.

Referring to FIG. 2, the automatic article providing system of the present invention includes an article generating apparatus 100, an information provider 200, a customer terminal 400, an administrator terminal 500 connected to the article generating apparatus 100, and a communication network. And 300.

The information provider 200 includes, for example, stock exchanges, KOSDAQ, the Financial Supervisory Service, and any other institution that provides formal information related to securities.

The information posted by the information provider 200 is processed and monitored by the article generating apparatus 100 through the information communication network 300 and provided to the customer terminal 400.

The customer terminal 400 is a personal computer or news provider connected to the communication network 300.

The article generating apparatus 100, the information provider 200, and the customer terminal 400 are connected to a communication network 300 that can be accessed in a wired or wireless manner through the Internet or a LAN, and the customer terminal 400 is connected to the Internet. Any terminal can be connected.

The article generating apparatus 100 monitors the information posted by the information provider 200 in real time, determines whether the information is valid, and provides the article to the customer terminal through a predetermined process, and the customer terminal 400 provides the news. In the case of businesses, the delivered articles are delivered to individuals who have access to the Internet through a personal computer (PC).

Referring to FIG. 3, the article generating apparatus 100 of the present invention includes a hardware including a central processing unit, a RAM, an external storage device, a communication module, and software executed thereon. The storage unit 120, the data processing unit 140, and the article generation unit 160 are configured to be included.

The storage unit 120 includes a search word DB 121, a metadata DB 123, a cumulative data DB 125, a processed data DB 127, and a stock quote data DB 129. The data processor 140 includes a timer 141, a search and classification module 143, a parsing and validation module 145, and a processing module 147. The news generator 160 includes an analysis module ( 162, tabulation module 164, and thesaurus DB 168.

Hereinafter, the article generating apparatus 100 of the present invention will be described in detail.

Timer 141 of the data processing unit 140 is linked to the search and classification module 143, the search and classification module 143 in real time or periodically connected to the server of the information provider 200 to the disclosure information Provides time information needed to search for

The disclosure information includes, for example, performance disclosures related to corporate performance, order disclosures related to orders, equity disclosures related to equity changes, and other broad types of disclosures. An example will be described.

The search and classification module 143 connects to the server of the information provider 200 through the communication network at a time set by the timer 141 or searches the data provided from the information provider 200 in real time or periodically to index the list. When a match is generated with the generated metadata (Meta data) containing the announcement and the unique number and stored in the metadata DB (123). The search and classification module 143 searches the data by using the index word list stored in the search word DB 121. An example of the index word list stored in the search word DB 121 is shown at 510 in FIG. 5. An example of the search and classification result of the data is shown by reference numeral 520 of FIG. 5.

The parsing and validation module 145 parses the contents of the searched and classified data, and if there is missing data during parsing, it records as "missing" whether or not the metadata accumulation data is generated and judges it as invalid data. Solution The unique number is terminated without any further processing. When parsing starts, if it is determined that the disclosure information is not worth accumulating (reference numeral 705 of FIG. 7), data of raw disclosure text is expressed in a tabular form as it is, as indicated by reference numeral 715 of FIG. 7. The current performance data 725, which is used to store and accumulate, is stored in the cumulative data DB 125 as shown by reference numeral 735 of FIG. In this case, it is recorded as "in inspection" in the accumulated data validity column 610 in the metadata of FIG. 6.

In addition, the parsing and validation module 145 determines whether there is a description error with respect to the parsed data without omission, and corresponds to "before inspection" in the "cumulative data validity column 610" item of FIG. The inspection logic for the disclosure information is performed by referring to the unique number of the item, and the inspection logic is different according to the disclosure information.

For example, the inspection logic of performance disclosure among the disclosure information includes longitudinal inspection and horizontal inspection of financial statements, and the horizontal inspection is based on the assumption that the company's sales rarely change more than the previous quarter. If it is out of a certain range set by, it is recognized as an error. This error is often caused by mis-entering the unit when the company enters the disclosure (eg, when it is stated between 100 million won and 1 million won).

Although the news may not be generated for a company whose sales are abnormally changed by the horizontal method, the present invention prioritizes the purpose of preventing wrong information and thus ends the method described below. Longitudinal inspection is to check whether the value of the deduction between items using accounting principles is consistent.

The processing module 147 performs processing on the disclosure information by referring to the unique number of the item that is recorded as valid in the "cumulative data validity column 610" stored in the metadata DB 123.

At this time, if the past performance is quarterly performance, the processing module 147 calculates processed data such as annualized financial ratio through the past quarterly performance linkage process by referring to the past quarterly performance of the cumulative data DB 125 of the corresponding item. When the data is processed immediately without interlocking with the cumulative data DB (125). Regarding the processed data, it is recorded in the metadata of FIG. 6 in the processing data generation column 620 item as "complete".

The processed data is data including at least one of a stock price return ratio (PER), an operating margin and a return on equity (ROE) used in the investment information.

In addition, the processing module 147 generates the processing data through interworking with different DBs rather than the same DB. For example, in order to obtain a return on equity, the net profit data of FIG. References to the total equity items in the balance sheet of data accumulated by parsing in other disclosures.

The processing data is typically useful as investment information for judging business performance of a time-series company or comparing it with other companies, and examples of phrases and graphs generated from the data are shown in FIG. 8.

As described above, the data processed by the data processing unit 140 is referred to by the article generation unit 160.

The analysis module 162 includes functions of time series analysis and inter-company comparison, and "individual logic" is set in the analysis module 162 for analysis according to each situation, and the processing of the result is analyzed module. "Universal logic" at 162 refers to thesaurus DB 168 to generate the final sentence. Here, the "individual logic" is inputted by the administrator through the administrator terminal 500, and is a logic for obtaining one of the plurality of analysis information (H) according to the combination of the talk and the verb, and the "general logic" is the "individual logic". It is the logic to automatically process the survey and the mother in response to the result composed of the message and the verb obtained by ". It is set at the initial configuration of the system, and then goes through a simple update management process. 9 is a diagram illustrating an embodiment in which the analysis module 162 generates a sentence by analyzing simple logic.

The analysis module 162 analyzes the data of the cumulative data DB 125 or the data of the processed data DB 127 by inputting " individual logic " in advance and shows the result as shown in the output of FIG. As for the data at the output terminal, the "general logic" of the analysis module 162 refers to the sentence generation thesaurus DB 168 to generate a sentence suitable for each situation.

The tabulation module 164 generates graphs and / or tables by referring to the accumulated data 125 and the data of the processed data DB 127 without any separate data processing. The tabulation module 164 generates a graph and / or a table of the corresponding item by referring to the item of the unique number whose cumulative data on the metadata of FIG. 6 is "valid" and the processed data is "complete", and generates by the above process. Create articles with printed sentences.

11 is an exemplary view of parsing the original document of the order disclosure through the article generating device 100 of the present invention to the article relating to the order disclosure.

The data processing unit 140 of the article generating apparatus 100 parses the core content of the order disclosure, that is, the disclosure data (that is, the contracted goods, the contracted amount, the recent sales, the percentage of the sales, and the counterpart) T10. First, the processed or analyzed opinions processed by the data processing unit 140 based on the published data T10 are converted and displayed by converting the disclosure data (T10) included in the original text of the order to the article (T20) in the sentence form. (T30) is displayed.

In addition, reference numeral T40 indicates the cumulative disclosure data generated by listing data obtained from the cumulative data DB 125 in which the disclosure texts such as the disclosure data T10 are accumulated, and the reference number T50 provides information. It is part of a stock price data article created in conjunction with separate data, such as stock quote data DB 129, that is not associated with data provided from the circle 200.

As described above, in the present invention, even in the case of an order disclosure, the processing or analysis opinion (T30), the cumulative disclosure data (T40), and the stock price data article (T50), as well as the article (T20) in sentence form based on the disclosure data (T10). It is possible to provide at the same time.

Therefore, in the present invention, the user is informed of analytical opinions and the like by linking the base data, the historical cumulative data, and the processed data, and displays all the data to be searched by the user. It can be used as.

The operation of the article generating device of the present invention configured as described above will be described in detail with reference to FIG.

4 is a flowchart illustrating a method for automatically generating articles according to an embodiment of the present invention.

Prior to operating the article generating apparatus 100, the administrator accesses the article generating apparatus 100 through the manager terminal 500 and inputs a search term index, a data platform definition, for determining whether a publication to be published is published or not. It performs sentence generation thesaurus, timer setting, meta data definition, and individual module insertion in each module. After the above-mentioned input work is completed, an article is automatically generated whenever a publication occurs for the corresponding disclosure. The manager terminal 500 may be connected to the article generating apparatus 100 to change or delete the above-described input work.

The timer 141 of the article generating apparatus 100 may search and classify the data provided from the information provider 200 connected through the communication network 300 based on a predetermined index word and classify the searched data when the predetermined time is reached. Operation 143 (S400).

The search and classification module 143 is operated in real time to generate articles requiring breaking news, and searches for a list of documents that require real time.

In addition, the retrieval and classification module 143 operates with a predetermined period for articles that are not urgently required or need to collect and process and report a large amount of data, and mainly operate at a time when the stock market is not opened. .

In an application example, the timer 141 may be configured for the search and classification module 143 only during office hours during which disclosures are likely to occur (eg, from 9 am to 6 pm, or from 9 am to 3 pm, etc.). ) Is set to operate in real time, it is possible to set the search and classification module 143 to operate with a predetermined period when the working time is exceeded.

The search and classification module 143 searches the server of the information provider 200 in real time (for example, for 5 seconds) or periodically to generate an item that matches the list on the index. Generate the metadata containing the stored in the metadata DB (123) (S410).

It divides the classification of disclosure and decides what kind of article will be created in the later process, and the unique number prevents duplication of the same contents. 6 illustrates an example of the generated metadata. When the metadata is generated and stored in the metadata DB 120 for the first time, it is recorded as “first” in the accumulated data validity column 610.

The parsing and validation module 145 parses the contents of the searched and classified data, and if there is missing data during parsing, it records as "missing" whether or not the metadata accumulation data is generated and judges it as invalid data. Solution The unique number is terminated without any further processing.

When there is no missing data as a result of parsing the content in the parsing and validating module 145, the data (725 and 735 of FIG. 7) that are worth accumulating among the data are unified (eg, 100 million won) and stored in the cumulative data DB (125) as the third quarter performance data (735) (S420). In the metadata of FIG. 6, it is recorded in the column 610 whether cumulative data is generated.

In addition, the parsing and validity checking module 145 determines whether there is a description error with respect to the parsed data without omission (S430).

The parsing and validation module 145 performs inspection logic on the disclosure information by referring to a unique number of an item corresponding to "before inspection" in the "cumulative data validity column 610" item in the metadata of FIG. 6. The inspection logic is different according to the disclosure information as described above.

When an error occurs in the horizontal inspection or longitudinal inspection, the disclosure officer of the company reports that the data misrepresentation error occurred during the disclosure, and the parsing and validation module 145 checks the "cumulative data generation column 610" of metadata. "Error" is recorded (not shown) and is deleted by referring to the year and month of the data in the cumulative data DB (125) and does not process subsequent processes for the unique number. The parsing and validity checking module 145 writes (valid) to the " cumulative data generation column 610 "

The processing module 147 performs the processing on the disclosure information by referring to the unique number of the item recorded as valid in the "cumulative data validity column 610" stored in the metadata DB 123, as shown in FIG. Process data is generated (S440).

At this time, if the past performance is quarterly performance, the processing module 147 calculates processed data such as annualized financial ratio through the past quarterly performance linkage process by referring to the past quarterly performance of the cumulative data DB 125 of the corresponding item. When the data is processed immediately without interlocking with the cumulative data DB (125). Regarding the completed data, the state "before generation" is recorded as "complete" in the processing data generation column 620 in the metadata of FIG. 6.

The processed data generated in this way is referred to by the analysis module 162 and the tabulation module 164 of the article generator 160.

The analysis module 162 of the article generator 160 includes functions of time series analysis and intercompany comparison, and the analysis according to each situation is performed by “individual logic” in the analysis module 162. Investigation of the result and processing of the mother generate the final sentence in such a manner that "general logic" of the analysis module 162 converts numerical data into sentence data with reference to the thesaurus DB 168 (S450).

The analysis module 162 analyzes the data of the cumulative data DB 125 or the data of the processed data DB 127 by pre-input "individual logic" and shows the analysis result as shown in the output stage of the flowchart of FIG. 9. The analysis data (901-904) at the output terminal generates sentences for each situation with reference to the sentence generation thesaurus DB (168), and the analysis module (162) automatically searches for the word and the message, etc. To be processed.

The flowchart of FIG. 9 shows, for example, a total of four results by judging whether the quarterly sales and operating profits decrease or increase, respectively, and FIG. 10 is a diagram showing the internal structure of a sentence generation thesaurus DB. And explanatory drawing which shows the process of obtaining analysis information.

First, for the statement defined as sales (A), if the sales increase according to whether the sales increase or decrease by referring to the data item from the DB, the proverb is determined to be C1, and if the sales decreases, the proverb is determined to be C2. In the same way as described above, the wording is determined as G1 when increasing with respect to the operating profit E, and as G2 when decreasing.

When C1 and G1 are determined, the analysis information (H) is determined as "coming boom" (H1), when C1 and G2 is determined, the analysis information (H) is determined as "business quality improvement" (H2), C2 and When G1 is determined, analysis information H is determined as "profit margin improvement" (H3), and when C2 and G2 are determined, analysis information H is determined as "coming recession phase" (H4).

For example, if the sales increase and the operating profit increase, the result of A1, C1, E1, G1, H1 is derived. In other words, the word (C) and the word (G) for the word (A) and the word (E). It is the role of "individual logic" to derive the result analysis information (H) by applying. Also, for the final results A1, C1, E1, G1, and H1, the surveys (b1, f1) are applied between the verbs (A, E) and the verbs (C, G), respectively. d1) is the "general purpose logic".

As a result, when the sales increase and the operating profit increases as described above, "the amount of sales increased and the operating profit also increased from the numerical data by the application of individual logic and general-purpose logic according to the thesaurus for generating sentences of the present invention. A sentence such as "" is automatically generated, and as a result, the analysis information "H" of "coming active stage" is derived. That is, using the sentence generation thesaurus of the present invention, natural sentences are generated by sequential progress of individual logic and general purpose logic.

As described above, in the present invention, after extracting the statement from the numerical data, selecting the word according to the increase or decrease of the numerical data, determining the analysis information according to the combination of the selected word and the word, and then applying a predetermined rule between the word and the word. It is possible to convert the numerical data into sentence data by inserting the survey and the ending.

Further, after converting the numerical data into sentence data in the analysis module 162, or referring to the cumulative data DB 125 and the processing data DB 127 irrespective of the step S450, / Or generates a graph (S460).

The article data is generated by combining the sentence data generated in step S450 and the table and / or graph generated in step S460 and provided to the customer terminal 400 through a communication network (S470).

In the above embodiment, the Financial Supervisory Service disclosure, electronic disclosure, etc., which are referred to for stock trading, are described as examples. However, the present invention relates to real estate prices, weather information, life insurance association product information, and asset management association funds. Application to information and the like is also possible by those skilled in the art.

As described in detail above, according to the method and apparatus for automatically generating articles of the present invention, in addition to abundant additional data and analytical opinions for the information user's understanding in generating news related to company disclosure, the entire process may be automated. You can provide articles with real-time concepts that are more accurate and faster than manual editing.

In addition, the present invention informs the analysis opinion through the linkage of the base data, the historical cumulative data and the processed data, and displays all the data to be searched by the user so that they can be utilized as a basis for rich and comprehensive opinions or investment decisions. Provide an analysis article.

In the article generating apparatus of the present invention, all disclosure information is converted into articles for all disclosure information within the disclosure, and the arbitrary processing is excluded, and the article processed by the apparatus is disclosed to all clients equally so that information asymmetry does not occur. There is a point.

In addition, in the conventional news generation system, the presence or absence of an event corresponding to logic is an important determinant and content of news generation, and the content to be delivered is an occurrence of an event corresponding to logic. In the present invention, however, the present invention is faithful to the reporting principle of the article and provides abundant contents because of the purpose of providing additional data and analyzing the incident.

Although the present invention has been described in detail with reference to the preferred embodiments, the present invention is not limited to the above-described embodiments, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the following claims. Anyone skilled in the art will have the technical idea of the present invention to the extent that various modifications or changes are possible.

Claims (10)

a) retrieving, collecting and classifying data provided from information sources connected through a communication network; b) generating metadata about the data provided from the information provider and storing the metadata in the metadata DB; c) parsing the contents of the data provided from the information provider, determining whether there is missing data and recording it as "missing" in the corresponding item of metadata; d) extracting and storing the cumulative data worth accumulating among the parsing contents of the data provided from the information provider if the missing data does not exist as a result of the determination of step c); e) generating processing data used for article generation based on the accumulated data; And f) generating an article by converting the numerical data contained in the processed data into sentence data. The method of claim 1, wherein in step d) And when the parsing data is accumulated and stored as a result of parsing the data provided from the information provider, recording a predetermined mark on whether to generate the cumulative data of the metadata. The method of claim 1, wherein in step d) Checking whether there is an error in the description of the parsing content, and if there is no error in the description step e) characterized in that the automatic generation of articles. The method of claim 1, wherein in step e), And when the cumulative data is quarterly performance data, generating processing data in association with past quarterly earnings. The method of claim 4, wherein the processing data, And at least one of a stock price return ratio, an operating profit margin, and a return on equity. The method of claim 1, wherein step f) Extracting a spoken word from the numerical data, selecting a spoken word according to the increase and decrease of the numerical data, and determining analysis information according to a combination of the selected spoken word and the spoken word; And And inserting a search and a parent according to a predetermined rule between the spoken word and the used word to complete a sentence. A search and classification module for searching, collecting, and classifying data generated from an information provider providing at least one disclosure information connected to a communication network based on a given index word; A parsing and validation module for parsing desired content from data generated from the retrieved information provider and validating the parsed data; A data processing unit including a processing module configured to generate processing data used for article generation based on the parsed data on which the validity check is performed; And And an article generation unit for generating an article by converting numerical data included in the processing data provided from the data processing unit into sentence data. The method of claim 7, wherein the data processing unit, And an accumulating data DB for accumulating and storing some of the parsed data and a machining data DB for storing machining data processed by the machining module. The method of claim 8, wherein the article generating unit, And a tabulation module for generating at least one of a table and a graph based on the data of the cumulative data DB and the processed data DB. The article generating unit according to any one of claims 7 to 9, After extracting the statement from the numerical data, selecting the term according to the increase and decrease of the numerical data, determining the analysis information according to the combination of the selected term and the term, and then searching and ending according to a preset rule between the term and the term. And automatically inserting the numerical data into sentence data by inserting the article.

Images