KR102468975B1 - Method and apparatus for improving accuracy of recognition of precedent based on artificial intelligence - Google Patents

Abstract

Provided are a method and device for improving accuracy of case precedent recognition based on artificial intelligence. The method may include the steps of: generating multiple types of optical character recognition (OCR) data by processing original data containing the case precedent contents; extracting errors in the content of the case precedent using the multiple types of OCR data; and correcting the extracted errors.

Description

Artificial intelligence-based case law recognition accuracy improvement method and apparatus

The present disclosure relates to a method and apparatus for improving the accuracy of case precedent recognition based on artificial intelligence.

Recently, with the rapid spread of digital storage media, digitization of documents that previously existed on paper has been actively developed. This phenomenon is accelerating further with the development of Optical Character Recognition (OCR), which is a technology for automatically recognizing characters included in documents.

Optical Character Recognition (OCR) is the acquisition of images of human-written or machine-printed characters with an image scanner and conversion into machine-readable characters.

On the other hand, it is very difficult to classify and organize judgments because, by their nature, there are not only a large number of existing judgments, but also new judgments are constantly created, and at the same time, there are cases in which the standard precedent is changed. Also, unless you read all the rulings, it is difficult to know whether this is an important ruling for you or whether it is necessary.

Registered Patent Publication No. 10-1516684, 2015.04.24.

The problem to be solved by the present disclosure is to provide a method and apparatus for improving the accuracy of case precedent recognition based on artificial intelligence.

However, the problem to be solved by the present disclosure is not limited to the above problem, and other problems may exist.

A method for improving the accuracy of artificial intelligence-based precedent recognition according to an aspect of the present disclosure for solving the above problems includes generating a plurality of OCR (Optical Character Recognition) data by processing original data containing precedent contents, Extracting an error in the content of the precedent using the plurality of OCR data and correcting the extracted error, wherein the error extraction step includes the plurality of errors based on the bidirectional context of the character to be verified. It may include extracting a first error by analyzing OCR data of .

In addition, the OCR data generating step may include generating OCR data corresponding to the number of times by repeating OCR on the original data a predetermined number of times.

In addition, in each of the plurality of OCR data, characters at the same position may be differently expressed according to the accuracy of each OCR performed repeatedly the number of times.

In addition, the accuracy may be calculated based on a ratio of a part recognized as a text and a part recognized as an image in the original data.

In addition, the first error extraction step compares each of the plurality of OCR data based on the bidirectional context of the character to be verified, and extracts a character having a difference between the plurality of OCR data based on the compared result. it may be

The error correction step may include setting a correction direction for the first error and applying the set correction direction to the first error.

In addition, the correction direction includes a first type of deleting an erroneous character, a second type of inserting a missing character, a third type of selecting one of a plurality of characters, and a new character after deleting the erroneous character and replacing the deleted character. A fourth type of insertion may be included.

An apparatus for improving the accuracy of artificial intelligence-based precedent recognition according to another aspect of the present disclosure for solving the above problems includes a communication unit, a memory for storing at least one process for improving the accuracy of artificial intelligence-based case precedent recognition, and the and a processor operating according to a process, wherein the processor generates a plurality of OCR (Optical Character Recognition) data by processing original data including precedent contents based on the process, and uses the plurality of OCR data. to extract an error in the contents of the precedent, correct the extracted error, and when extracting the error, the processor analyzes the plurality of OCR data based on the bidirectional context of the character to be verified to determine the first error can be extracted.

In addition, when generating the OCR data, the processor may repeat OCR on the original data a predetermined number of times to generate OCR data corresponding to the number of times.

In addition to this, another method for implementing the present disclosure, another system, and a computer readable recording medium recording a computer program for executing the method may be further provided.

Other specific details of the disclosure are included in the detailed description and drawings.

According to the present disclosure described above, errors generated during OCR recognition are extracted and corrected through comparative analysis of a plurality of OCR data obtained by performing OCR several times for precedents in the form of images, thereby providing users with more accurate text form. case law can be provided.

The effects of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a block diagram of an apparatus for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure.
2 is a flowchart of a method for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure.
3 is a flowchart of a specific method of step S120 of FIG. 2 .
4 is a diagram for explaining a first error according to the present disclosure.
5 is a diagram for explaining error correction according to the present disclosure.
6 is a diagram for explaining an artificial intelligence algorithm used for performing each step of FIG. 2 .

Like reference numbers designate like elements throughout this disclosure. The present disclosure does not describe all elements of the embodiments, and general content or overlapping content between the embodiments in the technical field to which the present disclosure belongs is omitted. The term 'unit, module, member, or block' used in the specification may be implemented as software or hardware, and according to embodiments, a plurality of 'units, modules, members, or blocks' may be implemented as one component, It is also possible that one 'part, module, member, block' includes a plurality of components.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case of being directly connected but also the case of being indirectly connected, and indirect connection includes being connected through a wireless communication network. do.

In addition, when a certain component is said to "include", this means that it may further include other components without excluding other components unless otherwise stated.

Throughout the specification, when a member is said to be located “on” another member, this includes not only a case where a member is in contact with another member, but also a case where another member exists between the two members.

Terms such as first and second are used to distinguish one component from another, and the components are not limited by the aforementioned terms.

Expressions in the singular number include plural expressions unless the context clearly dictates otherwise.

In each step, the identification code is used for convenience of description, and the identification code does not explain the order of each step, and each step may be performed in a different order from the specified order unless a specific order is clearly described in context. have.

Hereinafter, the working principle and embodiments of the present disclosure will be described with reference to the accompanying drawings.

In the present specification, the 'apparatus for improving the accuracy of recognizing artificial intelligence-based precedents according to the present disclosure' includes various devices capable of providing results to users by performing calculation processing. For example, the apparatus for improving the accuracy of case law recognition based on artificial intelligence according to the present disclosure may include a computer, a server device, and a portable terminal, or may be in any one form.

Here, the computer may include, for example, a laptop computer, a desktop computer, a laptop computer, a tablet PC, a slate PC, and the like equipped with a web browser.

The server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a web server.

The portable terminal is, for example, a wireless communication device that ensures portability and mobility, and includes a Personal Communication System (PCS), a Global System for Mobile communications (GSM), a Personal Digital Cellular (PDC), a Personal Handyphone System (PHS), and a PDA. (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), WiBro (Wireless Broadband Internet) terminal, smart phone ) and wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted-devices (HMDs). can include

1 is a block diagram of an apparatus for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure.

2 is a flowchart of a method for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure.

3 is a flowchart of a specific method of step S120 of FIG. 2 .

4 is a diagram for explaining a first error according to the present disclosure.

5 is a diagram for explaining error correction according to the present disclosure.

6 is a diagram for explaining an artificial intelligence algorithm used for performing each step of FIG. 2 .

The artificial intelligence-based case precedent recognition accuracy improvement device 100 (hereinafter referred to as the accuracy improvement device) of the present disclosure performs OCR to convert the image into text when obtaining case precedent data in the form of an image, and the OCR data may include It is possible to improve the accuracy of precedent recognition by extracting and correcting existing errors.

Referring to FIG. 1 , an accuracy improving device 100 may include a communication unit 110 , a memory 120 and a processor 130 .

The communication unit 110 may include one or more components enabling communication with an external device (not shown), for example, at least one of a wired communication module, a wireless communication module, a short-distance communication module, and a location information module. can include Here, the external device (not shown) may be a user's terminal and a server device, but is not limited thereto.

The memory 120 may store data supporting various functions of the accuracy improving device 100 and a program for the operation of the processor 130, and input/output data (eg, music files, stop images, videos, etc.), and can store a plurality of application programs (applications) running in the device, data for the operation of the accuracy improving device 100, and commands. At least some of these application programs may be downloaded from an external server through wireless communication.

More specifically, the memory 120 stores at least one process for improving the accuracy of AI-based case precedent recognition, so that the processor 130 improves the accuracy of AI-based precedent recognition of the present disclosure based on the process. method can be done.

The memory 120 may be a flash memory type, a hard disk type, a solid state disk type, a silicon disk drive type, or a multimedia card micro type. micro type), card type memory (eg SD or XD memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable EEPROM (EEPROM) It may include a storage medium of at least one type of a programmable read-only memory (PROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. In addition, the memory 120 is separated from the apparatus 100, but may be a database connected by wire or wirelessly.

In addition to operations related to the application program, the processor 130 may generally control overall operations of the accuracy improving device 100 . The processor 130 provides appropriate information or functions to an external device (not shown) by processing signals, data, information, etc. input or output through the components described above or by driving an application program stored in the memory 120, or can be dealt with

In addition, the processor 130 controls any one or a plurality of the components described above in order to implement various embodiments according to the present disclosure described in FIGS. 2 to 6 below on the accuracy improving device 100. can do.

Specifically, the processor 130 generates a plurality of OCR (Optical Character Recognition) data by processing the original data including the precedent contents, extracts an error in the precedent contents using the plurality of OCR data, and extracts an error in the precedent contents. errors can be corrected. At this time, when extracting the error, the processor 130 extracts a first error by analyzing the plurality of OCR data based on word order, and then extracts the plurality of OCR data based on a precedent-related expression pre-stored in the database. A second error may be extracted by analyzing the OCR data.

At least one component may be added or deleted corresponding to the performance of the components shown in FIG. 1 . In addition, it will be easily understood by those skilled in the art that the mutual positions of the components may be changed corresponding to the performance or structure of the system.

Meanwhile, each component shown in FIG. 1 means software and/or hardware components such as a Field Programmable Gate Array (FPGA) and an Application Specific Integrated Circuit (ASIC).

Hereinafter, with reference to FIGS. 2 to 6, a method for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure will be described in detail. Each operation of the method according to the present disclosure may be understood to be performed by the accuracy improving device 100, but is not limited thereto.

Referring to FIG. 2 , the processor 130 of the accuracy improving device 100 may generate a plurality of OCR (Optical Character Recognition) data by processing original data including the case law contents (S110).

Here, the original data may be received from an external device (not shown) through the communication unit 110 . In addition, the original data is precedent data including precedent contents, and may be data in the form of an image.

When a user who needs precedent data in text form provides the original data in the form of an image to the accuracy improving device 100, the accuracy improving device 100 performs OCR recognition on the original data to provide the user with the precedent in text form. can provide At this time, the accuracy improving apparatus 100 may perform OCR multiple times to improve the accuracy of OCR recognition, and provide more accurate content of a precedent to the user through comparative analysis of a plurality of OCR data as a result of OCR performance.

Specifically, the processor 130 may repeat OCR on the original data a predetermined number of times to generate OCR data corresponding to the number of times.

For example, if the preset number of times is three, the processor 130 may perform OCR on the same original data three times to generate first OCR data, second OCR data, and third OCR data.

In this case, in each of the plurality of OCR data, characters at the same location may be differently expressed according to the accuracy of each OCR performed repeatedly the number of times.

Here, the accuracy may be a value obtained by quantifying quality scores of each OCR performed on the original data. That is, when OCR is repeatedly performed on the same data, the accuracy of each OCR data may be different, and thus, the text included in each OCR data, that is, the recognized text may be different. For example, when performing the first OCR, a character at a specific position may be expressed as “this”, and when performing the second OCR, a character at the same position may be expressed as “0ㅣ”.

Depending on the embodiment, the accuracy may be calculated based on a ratio of a part recognized as a text and a part recognized as an image in the original data. However, the accuracy of each OCR data may be calculated based on various variables without being limited thereto. The higher the ratio of the part recognized as an image rather than text, the lower the accuracy of the corresponding OCR data.

Depending on the embodiment, only OCR data whose accuracy calculated from among a plurality of OCR data is equal to or greater than a preset value may be used as analysis target data. That is, data with too low accuracy may be discarded, and only data with accuracy higher than a certain standard may be used as analysis data for error extraction.

Then, the processor 130 of the accuracy improving device 100 may extract an error in the content of the precedent using the plurality of OCR data (S120).

Referring to FIG. 3 , the processor 130 may extract a first error by analyzing the plurality of OCR data based on the bi-directional context of the character to be verified (S121).

Here, the first error may mean an error generated in the OCR recognition process.

The processor 130 compares each of the plurality of OCR data based on the bidirectional context of the character to be verified, and extracts a word order in which a difference between the plurality of OCR data exists as a first error based on the compared result. can do.

Referring to FIG. 4, when two OCRs are performed on the sentence "The accused is sentenced to 6 months in prison. One year from the date of confirmation of this precedent..." among the contents of the precedent included in the original data, the first OCR result, the first OCR, is performed. The OCR data is created as “The accused is sentenced to 6 months in prison. 01 years from the final decision of the precedent…”, and the second OCR data, which is the result of the second OCR, is “The accused is sentenced to 6 months in prison. The processor 130 may determine the first error by analyzing all the letters located on the left and right of the letter to be verified using the first and second OCR data and calculating a probability value. For example, when the character to be verified is “6” in the OCR data shown in FIG. If the probability value calculated according to the bi-directional context is lower than a preset value, the processor 130 may determine that a first error exists with respect to the corresponding character to be verified.

Referring back to FIG. 3 , the processor 130 may extract a second error by analyzing the plurality of OCR data based on a precedent-related expression pre-stored in the database (S122).

Here, the second error may refer to a misspelling already included in the text of the precedent. That is, the accuracy improving apparatus 100 of the present disclosure can provide a more accurate result to the user by extracting not only errors generated during OCR recognition, but also characters that were originally written as misspelled from the original text even though the OCR recognition was successful.

The processor 130 may compare each of the plurality of OCR data with the precedent-related expression, and extract a phoneme-unit misspelling as a second error based on the comparison result.

Here, correct expressions and incorrect expressions (frequently used misspellings) for terms frequently used in court cases may be matched and stored in the database.

Depending on the embodiment, the processor 130 may analyze each of the plurality of OCR data using the database to determine whether a character stored in the database as an incorrect representation exists in the OCR data. Also, the processor 130 may extract the character determined to be an incorrect expression as a second error.

Depending on the embodiment, a correct expression and an incorrect expression for a specific word are matched and stored in the database. At this time, the incorrect expression may be stored by classifying the misspelling type according to the position of the syllable and phoneme of the corresponding word.

Specifically, among two specific words, an expression in which the first phoneme of the first syllable is incorrect is stored as the first misspelling of the word, an expression in which the second phoneme of the first syllable is incorrect is stored as the second misspelling of the word, and The expression with the third phoneme incorrect is stored as the third misspelling of the word, the expression with the incorrect first phoneme of the second syllable is stored as the fourth misspelling of the word, and the expression with the incorrect second phoneme of the second syllable is the fifth misspelling of the word. , and an expression in which the third phoneme of the second syllable is incorrect may be stored as the sixth misspelling of the corresponding word. However, the misspelling type of each word may be classified according to the number of syllables of each word and the number of phonemes of each syllable, without being limited thereto.

For example, in the database, "judgment" is stored as a correct expression, and as an incorrect expression matching it, "decision, approval" is stored as the first misspelling (an expression in which the first phoneme of the first syllable is incorrect), and "pun-gyeol, pin-gyeol" are stored in the database. ”is stored as the 2nd misspelling (incorrect expression of the second phoneme of the first syllable), “Panggyeol, Pamgyeol” is stored as the 3rd misspelling (incorrect expression of the 3rd phoneme of the first syllable), and “pannyeol, pandyeon” is the 4th misspelling. It is stored as an incorrect expression (the first phoneme of the second syllable is incorrect), “Pangeol, Pangyal” is stored as the fifth incorrect expression (the second phoneme of the second syllable is incorrectly expressed), and “Pangyeong, Pan??” is the sixth incorrect expression ( The third phoneme of the second syllable may be stored as an incorrect expression).

Referring back to FIG. 2 , the processor 130 of the accuracy improving device 100 may correct the extracted error (S130).

The processor 130 may set correction directions for the first error and the second error, and apply the set correction direction to the first error and the second error.

Depending on the embodiment, the correction direction may include a first type, a second type, a third type, a fourth type, and a fifth type.

The first type (DELETE) may be correction of deleting an erroneous character.

The second type (INSERT) may be a correction that inserts missing characters.

The third type (SELECT) may be correction for selecting one of a plurality of characters.

A fourth type (GENERATE (replace)) may be a correction in which an erroneous character is deleted and a new character is inserted in the deleted position.

In the fifth type (NO EDIT (remain)), if a plurality of OCR data are identical to each other, the word order that does not require correction is determined, and the existing data is maintained without performing a correction operation to change the recognition data (OCR data). have.

FIG. 5 illustrates a correction method for errors extracted from the first OCR data and the second OCR data shown in FIG. 4 .

In this case, error correction may be performed for each syllable. For example, as shown in FIG. 4, for each syllable, that is, “Phi”, “Go”, “In”,…, Calibration can be performed for each of “1”, “Year” and “Interval”.

Referring to FIG. 5 , a fifth type (NO EDIT (remain)) correction may be performed for “blood” and “high” where there is no difference between the first and second OCR data.

In the first OCR data, the first type (DELETE) correction may be performed for the “reverse” in which the same syllable is duplicated.

For the difference between the first and second OCR data, “wife” and “chi”, a third type (SELECT) correction of selecting (select 1) the correct “wife” out of the two can be performed. .

For “0” and “this”, which are the parts where there is a difference between the first and second OCR data, a third type (SELECT) of correcting “this” among the two can be selected (select 2). .

The first type (DELETE) correction may be performed on “1”, which is an unnecessary expression in the first OCR data.

A third type (SELECT) correction of selecting (select 2) the correct “one” out of the two for “this” and “one” that is a difference between the first and second OCR data may be performed. .

For “ ” (blank) and “1”, which are the differences between the first and second OCR data, a third type (SELECT) correction of selecting (select 2) the correct “1” out of the two can be performed. can

Hereinafter, with reference to FIG. 6, an artificial intelligence algorithm used when performing each step (S110 to S130) of the method for improving the accuracy of case precedent recognition based on artificial intelligence according to the present disclosure will be described.

In step S110, the processor 130 may perform OCR on the received precedent original data a plurality of times, and perform sequence aligner on a plurality of OCR data generated as a result of the execution.

In step S120, the processor 130 may perform Embedding, Self-Attention, and Feed Forward on the sequence-aligned data. At the same time, the processor 130 may perform Image Embedding on the precedent original data, and perform Feed Forward by cross-Attentioning the resultant data and the Self Attention data.

In step S130, the data obtained by performing Embedding, Self-Attention, and Feed Forward on the sequence-aligned data is input to the tag interpreter through a character generator, image embedding is performed on the original data of the precedent, and the result is performed. Cross-Attention of the data and the Self Attention data to perform Feed Forward, the resulting data can be input to the Tag interpreter through Character Tagger. And, tag interpreter text data can be output.

Unlike existing models, the transformer used in the embodiment of the present disclosure may use attention without using RNN or CNN.

The accuracy improving apparatus 100 according to an embodiment of the present disclosure enables more parallelization compared to existing models by using a transformer, and shows performance improvement that significantly reduces learning time.

2 and 4 describe that steps S110 to S130 are sequentially executed, but this is only an example of the technical idea of this embodiment, and those with ordinary knowledge in the art to which this embodiment belongs Since it will be possible to change and execute the order described in FIGS. 2 and 4 without departing from the essential characteristics of the present embodiment or to perform steps S110 to S130 in parallel, it will be possible to apply various modifications and variations, so FIG. 2 and FIG. 4 is not limited to a time-series order.

Meanwhile, in the above description, steps S110 to S130 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present disclosure. Also, some steps may be omitted if necessary, and the order of steps may be changed.

Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing instructions executable by a computer. Instructions may be stored in the form of program codes, and when executed by a processor, create program modules to perform operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

Computer-readable recording media include all types of recording media in which instructions that can be decoded by a computer are stored. For example, there may be read only memory (ROM), random access memory (RAM), magnetic tape, magnetic disk, flash memory, optical data storage device, and the like.

As above, the disclosed embodiments have been described with reference to the accompanying drawings. Those skilled in the art to which the present disclosure belongs will understand that the present disclosure may be implemented in a form different from the disclosed embodiments without changing the technical spirit or essential features of the present disclosure. The disclosed embodiments are illustrative and should not be construed as limiting.

100: accuracy improving device
110: communication department
120: memory
130: processor

Claims (10)

In the method performed by the device,
Generating a plurality of Optical Character Recognition (OCR) data by processing the original data including the contents of the precedent;
extracting errors in the content of the precedent using the plurality of OCR data; and
Correcting the extracted error based on a preset correction direction; Including,
The error extraction step,
extracting a first error by analyzing the plurality of OCR data based on the bi-directional context of the character to be verified; and
Extracting a second error by analyzing the plurality of OCR data based on an expression related to a judgment pre-stored in a database; including,
The OCR data generation step,
OCR is repeated for the original data a preset number of times to generate OCR data corresponding to the number of times, but only data with accuracy greater than a preset value among the plurality of OCR data is analyzed for error extraction and correction. used as data
The accuracy is calculated based on a ratio of a part recognized as a character and a part recognized as an image other than text in the original data,
The first error is an error generated when recognizing the OCR,
The second error is an error described in the precedent content itself included in the original data,
The error correction step is
Setting the correction direction for the first error and the second error, and applying the set correction direction to the first error and the second error;
The correction direction for the first error is set based on a result of comparing each of the plurality of OCR data based on the bidirectional context of the character to be verified and extracting a character having a difference between the plurality of OCR data, A first type of deleting an erroneous character at the same position in each of the plurality of OCR data, a second type of inserting a missing character at the same position in each of the plurality of OCR data, and the same position in each of the plurality of OCR data A third type of selecting a correct character from among a plurality of characters for and a fourth type of deleting an incorrect character for the same position in each of the plurality of OCR data and inserting a new character in the deleted position,
A method for improving the accuracy of case precedent recognition based on artificial intelligence. According to claim 1,
The second error extraction step,
Comparing each of the plurality of OCR data with the expression related to the sentence, and extracting different misspellings for each phoneme unit based on the compared result,
A method for improving the accuracy of case precedent recognition based on artificial intelligence. According to claim 2,
The misspellings include a first misspelling in which the first phoneme of the first syllable of a two-letter word is incorrect, a second misspelling in which the second phoneme of the first syllable is incorrect, a third misspelling in which the third phoneme of the first syllable is incorrect, and the second syllable of the word. Including a fourth error in which the first phoneme is incorrect, a fifth error in which the second phoneme of the second syllable is incorrect, and a sixth error in which the third phoneme of the second syllable is incorrect.
A method for improving the accuracy of case precedent recognition based on artificial intelligence. delete delete delete delete A program stored in a computer readable recording medium to be combined with a computer to execute the method of any one of claims 1, 2 and 3. communications department;
A memory storing at least one process for improving the accuracy of artificial intelligence-based precedent recognition; and
A processor operating according to the process; includes,
The processor, based on the process,
A plurality of OCR (Optical Character Recognition) data is generated by processing original data including precedent content, errors in the precedent content are extracted using the plurality of OCR data, and based on a preset correction direction, the extracted correct errors,
When the processor extracts the error,
Extracting a first error by analyzing the plurality of OCR data based on the bidirectional context of the character to be verified,
Extracting a second error by analyzing the plurality of OCR data based on an expression related to the judgment pre-stored in the database;
When the processor generates OCR data,
OCR is repeated for the original data a preset number of times to generate OCR data corresponding to the number of times, but only data with accuracy greater than a preset value among the plurality of OCR data is analyzed for error extraction and correction. used as data
The accuracy is calculated based on a ratio of a part recognized as a character and a part recognized as an image other than text in the original data,
The first error is an error generated when recognizing the OCR,
The second error is an error described in the precedent content itself included in the original data,
When correcting errors, the processor
Setting the correction direction for the first error and the second error, and applying the set correction direction to the first error and the second error;
The correction direction for the first error is set based on a result of comparing each of the plurality of OCR data based on the bidirectional context of the character to be verified and extracting a character having a difference between the plurality of OCR data, A first type of deleting an erroneous character at the same position in each of the plurality of OCR data, a second type of inserting a missing character at the same position in each of the plurality of OCR data, and the same position in each of the plurality of OCR data A third type of selecting a correct character from among a plurality of characters for and a fourth type of deleting an incorrect character for the same position in each of the plurality of OCR data and inserting a new character in the deleted position,
AI-based precedent recognition accuracy improvement device. According to claim 9,
When the processor extracts the second error,
Comparing each of the plurality of OCR data with the expression related to the sentence, and extracting different misspellings for each phoneme unit based on the compared result,
The misspellings include a first misspelling in which the first phoneme of the first syllable of a two-letter word is incorrect, a second misspelling in which the second phoneme of the first syllable is incorrect, a third misspelling in which the third phoneme of the first syllable is incorrect, and a second misspelling in the second syllable of the word. Including a fourth error in which the first phoneme is incorrect, a fifth error in which the second phoneme of the second syllable is incorrect, and a sixth error in which the third phoneme of the second syllable is incorrect.
AI-based precedent recognition accuracy improvement device.

Images