WO2021174823A1

WO2021174823A1 - Grammatical error correction method, apparatus, computer system, and readable storage medium

Info

Publication number: WO2021174823A1
Application number: PCT/CN2020/118197
Authority: WO
Inventors: 金晓辉; 阮晓雯; 徐亮
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-07-30
Filing date: 2020-09-27
Publication date: 2021-09-10
Also published as: CN111897535A

Abstract

Provided are a grammatical error correction method, apparatus, computer system, and readable storage medium, relating to the field of intelligent decision-making in artificial intelligence, comprising: obtaining an initial text, and inserting an actionable smart cursor at a preset position of said initial text (S100); performing real-time status marking of the initial text having a smart cursor to obtain real-time status information (S200); according to the real-time status information, using an error correction model to determine action data of the smart cursor (S300); using the smart cursor to process the initial text on the basis of the action data to obtain a target text (S400). The invention solves the problem in the prior art that grammatical error correction can be based only on limited predefined rules and mapping functions, it being time-consuming and the efficiency of error correction being low.

Description

Grammar error correction method, device, computer system and readable storage medium

This application claims the priority of a Chinese patent application filed on July 30, 2020 with the Chinese Patent Office application number 202010752813.1, titled "Syntax Error Correction Method, Device, Computer System, and Readable Storage Medium", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the field of intelligent decision-making in artificial intelligence, and in particular to a method, device, computer system, and readable storage medium for grammatical error correction.

Background technique

Before the program is run, it is often necessary to use a compiler to compile the program. If there are syntax errors in the written script, the compiler will report an error and cannot continue to run. In the prior art, most technicians rely on the compiler to obtain feedback on script error messages. However, the feedback information obtained in this way cannot accurately locate the specific grammatical error, which makes it very time-consuming to correct the grammatical error, and requires high professional requirements for the technicians.

The inventor found that in order to improve the efficiency of grammatical error correction, in the prior art, some common grammatical errors are pre-checked by means of preset rules or mapping, etc., and the check result is mapped to the correct grammatical code to correct the wrong grammar. However, the above-mentioned method to modify the grammar can only be based on limited predefined rules and mapping functions, and it takes a long time and the error correction efficiency is low.

technical problem

The purpose of this application is to provide a method, device, computer system, and readable storage medium for grammatical error correction, which can only be based on limited predefined rules and mapping functions, and is time-consuming. Long, the problem of low error correction efficiency.

Technical solutions

In order to achieve the above-mentioned purpose, this application provides a grammatical error correction method, including:

Acquiring the initial text, and inserting a smart cursor that can perform actions at a preset position of the initial text;

Mark the initial text with the smart cursor in real-time status to obtain real-time status information;

Using an error correction model to determine the action data of the smart cursor according to the real-time status information;

Using the smart cursor to process the initial text based on the action data to obtain the target text.

To achieve the above objective, this application also provides a grammar error correction device, including:

The preprocessing module is used to obtain the initial text, and insert a smart cursor that can perform actions at a preset position of the initial text;

The state acquisition module is used to mark the real-time state of the initial text with the smart cursor to obtain real-time state information;

An action determining module, configured to determine the action data of the smart cursor according to the real-time status information;

The action execution module is configured to use the smart cursor to process the initial text based on the action data to obtain the target text.

In order to achieve the foregoing objective, the present application also provides a computer device, the computer device including a memory, a processor, and a computer program stored in the memory and running on the processor. The processor executes the computer program when the computer program is executed. The following steps of the above grammatical error correction method:

To achieve the above objective, the present application also provides a computer-readable storage medium, which includes multiple storage media, each of which stores a computer program, and when the computer program stored in the multiple storage media is executed by a processor Jointly implement the following steps of the above grammatical error correction method:

Beneficial effect

The grammatical error correction method, device, computer system, and readable storage medium provided in this application insert an actionable smart cursor into the initial text, and determine the action data of the smart cursor through the error correction model based on the real-time status information of the initial text, Then perform actions on the initial text, and finally obtain the target text after the smart cursor completes all actions. The solution to the grammatical error correction existing in the prior art can only be based on limited predefined rules and mapping functions, and it takes a long time. The problem of low efficiency.

Description of the drawings

FIG. 1 is a flowchart of Embodiment 1 of the grammatical error correction method described in this application;

2 is a specific flow chart of real-time status marking of the initial text with smart cursor in the first embodiment of the grammatical error correction method described in this application to obtain real-time status information;

FIG. 3 is a specific flowchart of determining the action data of the smart cursor by using an error correction model according to the real-time status information in the first embodiment of the grammatical error correction method according to this application;

4 is a specific flowchart of training the error correction model before determining the action data of the smart cursor according to the real-time status information in the first embodiment of the grammatical error correction method of this application;

FIG. 5 is a specific flowchart of obtaining reward and punishment data according to the compilation result in Embodiment 1 of the grammatical error correction method according to this application;

FIG. 6 is a flowchart of using the smart cursor to process the initial text based on the action data in the first embodiment of the grammatical error correction method according to the application to obtain a target file;

FIG. 7 is a schematic diagram of program modules of Embodiment 2 of the enhanced grammatical error correction method according to this application;

FIG. 8 is a schematic diagram of the hardware structure of the computer device in the third embodiment of the computer system of this application.

Embodiments of the present invention

In order to make the purpose, technical solutions, and advantages of this application clearer and clearer, the following further describes the application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The grammatical error correction method, device, computer system, and readable storage medium provided in this application are suitable for the field of intelligent decision-making, and provide a grammatical error correction method based on a preprocessing module, a state acquisition module, an action determination module, and an action execution module . This application inserts a smart cursor that can perform actions into the initial text, determines the action data of the smart cursor through an error correction model based on the real-time status information of the initial text, and then executes actions on the initial text, and finally after the smart cursor completes all actions Obtain the target text, use the smart cursor as the smart carrier of the error correction model, locate the wrong grammar and modify it intelligently, and solve the existing grammatical error correction in the prior art, which can only be based on limited predefined rules and mapping functions, and is time-consuming Longer, the problem of lower error correction efficiency.

Example one

Please refer to Figure 1. A grammatical error correction method of this embodiment is used on the server side to automatically identify and correct grammatical errors before program coding, including the following steps:

S100: Obtain an initial text, and insert a smart cursor that can perform an action at a preset position of the initial text;

In this embodiment, the above-mentioned cursor is a flexible means for retrieving data from the table and performing operations. The cursor is mainly used on the server to process SQL statements sent from the client to the server, or batch processing, stored procedure, or trigger The advantage of the cursor is that it can locate a row in the result set and perform specific operations on the row of data. Specifically, the above-mentioned preset position is the head position of the initial text for subsequent use Smart cursor is used to correct grammatical errors from the head to the tail of the initial text.

S200: Mark the real-time status of the initial text with the smart cursor to obtain real-time status information;

Specifically, the above-mentioned real-time status mark is performed on the initial text with the smart cursor to obtain real-time status information, referring to Figure 2, including the following steps:

S210: Serialize the initial text to obtain first processed data;

In the foregoing embodiment, the serialization is in units of words, and the words include pre-defined functions, custom variables, operators and other units of the program. In the subsequent processing, the moving length of the smart cursor is also in words. The unit is the same as in the serialization process here.

S220: Locate the smart cursor in real time, obtain real-time position information of the smart cursor, mark the first processed data based on the position information, and obtain the first processed data with the cursor real-time position mark as real-time status information .

In this embodiment, the above-mentioned real-time status information is used to determine the specific position of the smart cursor in the initial file, so that the error correction model in the subsequent step S300 is used to automatically determine whether error correction is required according to the text at the position of the smart cursor. Actions (moving or editing) according to the text content in the forward direction of the position, the cursor position will continue to change, so the real-time status information can be obtained by marking the serialized initial text according to the real-time position of the cursor. Specifically, special text is used For example, <#cursor#> marks the position of the cursor, the initial position is the forefront of the initial text, and other special marks can also be used.

S300: Use an error correction model to determine the action data of the smart cursor according to the real-time status information;

It should be noted that the above action data includes two types: editing action data and navigation action data. The navigation action includes moving the position of the smart cursor in the initial text, which can move one word to the right or move down to the bottom of the initial text. The starting position of a sentence of code; editing actions include 3 types of insertion, deletion and replacement. The main editing object is defined as a variable set, including but not limited to semicolons, brackets, brackets, commas and dots. The above error correction model is LSTM The network combines the A2C model.

Specifically, the above-mentioned error correction model is used to determine the action data of the smart cursor according to the real-time status information, referring to FIG. 3, including the following steps:

S310: Use a neural network to perform mapping processing on the real-time status information to obtain first data;

In this embodiment, the LSTM neural network is used in step S310, and the real-time status information obtained after serialization in S211 is input into the long and short-term sequence (LSTM) network, and each word of the real-time status information is mapped to obtain a corresponding vector. The LSTM network is Used here as an encoding and decoding network.

S320: Perform element averaging processing on the first data to obtain second data;

In the above embodiment, the Mean Pooling layer (average pooling layer) is used to calculate the element average of the output vector to obtain the Embedding vector of the state, which is the second data. The above step S310 and step S320 process the real-time state information and transform it into the corresponding Vector.

S330: Use a deep reinforcement learning model to process the second data, and determine the smart cursor action data.

In this embodiment, the deep reinforcement learning model is an A2C model, and the A2C network is a multi-threaded reinforcement learning algorithm. Each thread contains its own thread network, which is divided into two parts: Actor network and Critic network. The Actor network is used to solve the action strategy, the Critic network is used to solve the value function, and the actor network is the input state (state) and the output action. Probability distribution, from which actions are selected as the input of the critic network; the critic network is to input state and action to estimate the q-value of the next state, and determine the action data of the smart cursor through the above-mentioned LSTM model combined with the A2C model.

In this embodiment, before determining the action data of the smart cursor according to the real-time status information, the error correction model is trained. Referring to FIG. 4, the training process includes the following:

S331: Obtain training samples, process the training samples by using a neural network, and obtain sample processing data;

Specifically, the training sample may be data similar to the initial text. As described in step S310 above, the training sample is processed by the LSTM neural network to obtain the corresponding sample vector.

S332: Use the action network and the state network in the deep reinforcement learning model to process the training sample processing data to obtain the initial action strategy and value function;

After the second data is input into the A2C model, that is, after the Embedding vector enters the thread, the linear layer 1 plus the softmax fully connected layer is used as the Actor to generate the action strategy; the linear layer 2 is used as the Critic to generate the value function.

S333: Use loss function processing to obtain sample action data based on the initial action strategy and value function;

In this embodiment, the loss function will adjust the initial action strategy and the value function to obtain the output sample action data, and the loss function will be adjusted during the training process.

S334: Use a compiler to compile the sample action data, and obtain reward and punishment data according to the compiling result;

In the training process, the error correction model is trained by the way of compiler feedback. Compared with some products in the industry with supervised learning and training, there is no need to give paired samples of the error code and the correct code at the same time, and there is no need to learn from In the process of manual modification and marking of samples, the rules of errors are sorted out, and the provision and selection of sample data is more flexible and convenient.

Referring to FIG. 5, in the above step S334, the reward and punishment data includes a positive punishment and a negative punishment. Specifically, obtaining the reward and punishment data according to the compilation result includes the following steps:

S334-1: Obtain the number of historical errors from the preset database, and obtain the number of errors after compilation according to the compilation result;

In the above embodiment, a preset database is provided for storing the number of errors after each compilation, and is updated according to the number of compilations to obtain the number of historical errors and the number of errors after compilation, which is mainly used to subsequently determine whether the errors of the compilation result increase or decrease, so that To determine whether it is a positive penalty or a negative penalty (ie reward).

S334-2: Determine whether the number of errors has increased based on the number of historical errors and the number of errors after compilation;

S334-3: If yes, the reward and punishment data is negative punishment;

S334-4: If not, the reward and punishment data is a positive punishment.

As an example and not limitation, the reward and punishment data in the database is updated based on the change in the amount of error data. If the amount of error data fed back in the compilation result increases, a negative penalty of -1 will be given; if the amount of feedback error data decreases, A positive reward of +1 will be given; if the compilation is passed, a positive reward of +100 will be given to end the iteration.

S334-4: After obtaining the penalty data, use the compiled error number to update the historical error number and store it in the preset database.

Specifically, the initial value of the reward and punishment data is preset, and then it will be updated according to the above-mentioned step S334-4, that is, the latest reward and punishment data is retained to adjust the above-mentioned initial data.

S335: Adjust the loss function and parameters in the error correction model based on the reward and punishment data, and process again until the training process is completed, and a trained error correction model is obtained.

Calculate the loss function according to the reward and punishment data, and update the parameters in the error correction model. The above S331-S334 and S335 adjust the loss function and parameters before it is a complete iterative process. Each iteration, each thread will use a synchronous update mechanism to adjust The data is passed to the global network.

S400: Use the smart cursor to process the initial text based on the action data to obtain the target text.

Specifically, the smart cursor is used to process the initial text based on the action data. Referring to FIG. 6, the process includes the following steps:

S410: Obtain a corresponding data type based on the action data;

As mentioned above, the data types include edit type and navigation type. The edit type is the need to modify the text at the location of the smart cursor, that is, the correction of incorrect grammar, and the navigation type is the guidance for the smart cursor , That is, the grammar is correct and does not need to be corrected, so that the smart cursor moves to the position of the next word.

S420: When the data type is an edit type, edit the initial text according to the action data, and update the real-time status information based on the position information of the smart cursor;

S430: When the data type is a navigation type, move the smart cursor according to the action data and update the real-time status information based on the position information of the smart cursor.

After using the smart cursor to process the initial text, whether it is editing or moving, it is finally necessary to update the real-time status information according to the position of the smart cursor, so as to determine whether the smart cursor executes or stops again according to the real-time status information .

After processing the initial text to obtain the target text, it also includes the following steps:

S440: Acquire current position information of the smart cursor, and determine whether the smart cursor is at the end of the initial text based on the position information;

S450: If yes, obtain the target text based on the processed initial text;

S460: If not, determine the action data of the smart cursor again according to the real-time status information.

In this embodiment, the special text <#cursor#> is used to mark the position of the cursor as described above. Therefore, it can be judged whether the cursor is at the end of the initial text according to the mark. If it is at the end of the initial text, it means that the cursor moves from the head of the initial text to the end. , Complete the grammatical error correction of the entire initial text. If it is not at the end of the initial text, repeat the above S310-S330 and S410-S440 based on the updated real-time status information to determine the execution action of the smart cursor again until the smart cursor reaches the end of the initial text .

It should be noted that, in order to further ensure the privacy and security of the above target text, the above target text can also be stored in a node of a blockchain, and the technical solution of this application can also be applied to other documents stored on the blockchain According to the classification, the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

In this solution, the smart cursor is used as the smart carrier of programming grammar error correction for intensive learning training, which can locate errors, intelligently modify them, and directly mobilize the compiler to make judgments after the modification during the training process. There is no need for technicians to manually troubleshoot errors. Modification and compilation solves the problem that the prior art can only be based on limited predefined rules and mapping functions and has low inspection efficiency.

Compared with the rule-based traversal search in the prior art, the smart cursor is given a series of actions that enable it to quickly reach the location that caused the program grammatical error, and directly make an action strategy based on the current text and cursor position information. higher efficiency.

Embodiment two:

Referring to FIG. 7, a syntax error correction device 5 of this embodiment includes:

The preprocessing module 51 is configured to obtain the initial text, and insert a smart cursor that can perform actions at a preset position of the initial text;

The status acquisition module 52 is configured to mark the initial text with the smart cursor in real-time status to obtain real-time status information;

The action determining module 53 is configured to determine the action data of the smart cursor according to the real-time status information;

It should be noted that the above action data includes two types: editing action data and navigation action data. The navigation action includes moving the position of the smart cursor in the initial text, which can move one word to the right or move down to the bottom of the initial text. The starting position of a sentence of code; editing actions include 3 types of insertion, deletion, and replacement. The main editing object is defined as a variable set, including but not limited to semicolons, brackets, brackets, commas, and dots.

The action determining module 53 includes the following:

The first processing unit 531 is configured to use a neural network to perform mapping processing on the real-time status information to obtain first data;

The neural network is an LSTM neural network.

The second processing unit 532 is configured to perform element averaging processing on the first data to obtain second data;

Specifically, use Mean The Pooling layer (average pooling layer) performs element averaging calculation on the output vector to obtain the Embedding vector of the state.

The third processing unit 533 is configured to use a deep reinforcement learning model to process the second data and determine the smart cursor action data.

The deep reinforcement learning model is an A2C model, and the A2C network is a multi-threaded reinforcement learning algorithm. Each thread will contain its own thread network, which is divided into two parts: Actor network and Critic network. The Actor network is used to solve the action strategy, and the Critic network is used to solve the value function. During the training process, the compiler is used to analyze the sample The action data is compiled, reward and punishment data is obtained according to the result of the compilation, the loss function and parameters in the error correction model are adjusted based on the reward and punishment data, and processed again until the training process is completed, and the error correction model is obtained.

The action execution module 54 is configured to use the smart cursor to process the initial text based on the action data to obtain the target text.

This technical solution is based on the detection model of intelligent decision-making, inserts an actionable smart cursor into the initial text through the preprocessing module, uses the state acquisition module to acquire the real-time state information of the initial text, and then uses the action determination module based on the acquired real-time state information Determine the action data of the smart cursor through the error correction model, and use the action execution module to use the smart cursor to perform actions on the initial text. Finally, the target text is obtained after the smart cursor completes all actions. The smart cursor is used as the smart carrier of the error correction model to locate Incorrect grammar and intelligently modify it, so as to solve the problem that grammatical error correction in the prior art can only be based on limited predefined rules and mapping functions, and takes a long time and has low error correction efficiency.

The technical solution is also based on the first processing unit, the second processing unit, and the third processing unit to determine the action data of the smart cursor through the error correction model. The error correction model is implemented through the LSTM model combined with the A2C model, and the smart cursor is controlled according to the action data. Editing to realize automatic error correction, and at the same time, after editing the text, a compiler will be used to compile the edited text, and the compilation result will be fed back to the adjustment error correction model to achieve independent learning of the error correction model and improve subsequent acquisition It is used for the accuracy of the action data of the smart cursor. At the same time, the error correction model does not need to give paired samples of the error code and the correct code at the same time. Training samples are provided and selected more flexibly and conveniently.

Embodiment three:

In order to achieve the above purpose, this application also provides a computer device 6 which includes multiple computer devices. The components of the grammatical error correction device 5 of the second embodiment can be dispersed in different computer devices. The device can be a smart phone, a tablet, a laptop, a desktop computer, a rack server, a blade server, a tower server, or a rack server (including a stand-alone server, or a server cluster composed of multiple servers) that executes the program Wait. The computer equipment in this embodiment at least includes but is not limited to: a memory 61 and a processor 62 that can be communicatively connected to each other through a system bus, as shown in FIG. 8. It should be pointed out that FIG. 8 only shows a computer device with components, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.

In this embodiment, the memory 61 (ie, readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), Read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disks, optical disks, etc. In some embodiments, the memory 61 may be an internal storage unit of a computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory 61 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), and a secure digital (Secure Digital, SD) equipped on the computer device. Card, Flash Card, etc. Of course, the memory 61 may also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory 61 is generally used to store an operating system and various application software installed in a computer device, such as the program code of the grammatical error correction method in the first embodiment, and so on. In addition, the memory 61 may also be used to temporarily store various types of data that have been output or will be output.

The processor 62 may be a central processing unit (Central Processing Unit) in some embodiments. Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 62 is generally used to control the overall operation of the computer equipment. In this embodiment, the processor 62 is configured to run program codes or process data stored in the memory 61, for example, to run a syntax error correction device, so as to implement the syntax error correction method of the first embodiment.

The network interface 63 may include a wireless network interface or a wired network interface, and the network interface 63 is generally used to establish a communication connection between the computer device 6 and other computer devices 6. For example, the network interface 63 is used to connect the computer device 6 to an external terminal through a network, and to establish a data transmission channel and a communication connection between the computer device 6 and the external terminal. The network may be an intranet (Intranet), the Internet (Internet), a global system of mobile communication (GSM), a wideband code division multiple access (WCDMA), 4G network, 5G Network, Bluetooth (Bluetooth), Wi-Fi and other wireless or wired networks.

It should be pointed out that FIG. 8 only shows the computer device 6 with components 61-63, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

In this embodiment, the grammatical error correction device 5 stored in the memory 61 may also be divided into one or more program modules, and the one or more program modules are stored in the memory 61 and are composed of one or more program modules. Multiple processors (the processor 62 in this embodiment) are executed to complete the application.

Embodiment four:

In order to achieve the above objective, the present application also provides a computer-readable storage medium. The computer-readable storage medium may be non-volatile or volatile, and includes multiple storage media, such as flash memory, hard disk, and multimedia. Card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), Programmable read-only memory (PROM), magnetic memory, magnetic disks, optical disks, servers, App application malls, etc., have computer programs stored thereon, and corresponding functions are realized when the programs are executed by the processor 62. The computer-readable storage medium of this embodiment is used to store a syntax error correction device, and when executed by the processor 62, the syntax error correction method of the first embodiment is implemented.

In an embodiment, the computer-readable storage medium includes a storage data area and a storage program area, the storage data area stores data created according to the use of blockchain nodes, and the storage program area stores computer programs; wherein When the computer program is executed by the processor 62, the grammatical error correction method described in any of the embodiments is implemented.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority or inferiority of the embodiments.

Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

A method for grammatical error correction, which includes:

Acquiring the initial text, and inserting a smart cursor that can perform actions at a preset position of the initial text;

Mark the initial text with the smart cursor in real-time status to obtain real-time status information;

Using an error correction model to determine the action data of the smart cursor according to the real-time status information;

Using the smart cursor to process the initial text based on the action data to obtain the target text.
The method for grammatical error correction according to claim 1, wherein the real-time status marking of the initial text with the smart cursor to obtain real-time status information includes the following:

Serialize the initial text to obtain first processed data;

The smart cursor is located in real time, real-time position information of the smart cursor is obtained, the first processed data is marked based on the position information, and the first processed data with the real-time position mark of the cursor is obtained as real-time status information.
The method for grammatical error correction according to claim 1, wherein the use of an error correction model to determine the action data of the smart cursor according to the real-time status information includes the following:

Use a neural network to perform mapping processing on the real-time status information to obtain the first data;

Performing element averaging processing on the first data to obtain second data;

A deep reinforcement learning model is used to process the second data to determine the smart cursor action data.
The method for grammatical error correction according to claim 1, wherein before determining the action data of the smart cursor according to the real-time status information, training the error correction model includes the following:

Obtaining training samples, processing the training samples by using a neural network, and obtaining sample processing data;

Use the action network and state network in the deep reinforcement learning model to process the training sample processing data to obtain the initial action strategy and value function;

Using loss function processing to obtain sample action data based on the initial action strategy and value function;

Use a compiler to compile the sample action data, and obtain reward and punishment data according to the compiling result;

The loss function and parameters in the error correction model are adjusted based on the reward and punishment data, and processed again until the training process is completed, and the trained error correction model is obtained.
The method for grammatical error correction according to claim 4, wherein said obtaining reward and punishment data according to the compilation result comprises the following:

Obtain the number of historical errors from the preset database, and obtain the number of errors after compilation according to the compilation result;

Judging whether the number of errors has increased based on the number of historical errors and the number of errors after compilation;

If it is, the reward and punishment data is a negative punishment; if not, the reward and punishment data is a positive punishment;

After the penalty data is obtained, the number of errors after compilation is used to update the number of historical errors and stored in the preset database.
The method for grammatical error correction according to claim 1, wherein said using said smart cursor to process said initial text based on said action data comprises the following steps:

Obtaining a corresponding data type based on the action data;

When the data type is edit data, edit the initial text according to the action data, and update the real-time status information based on the position information of the smart cursor;

When the data type is navigation data, move the smart cursor according to the action data and update the real-time status information based on the position information of the smart cursor.
The method for grammatical error correction according to claim 1, wherein after the initial text is processed and before the target text is obtained, the method further comprises the following steps:

Acquiring current position information of the smart cursor, and judging whether the smart cursor is at the end of the initial text based on the position information;

If yes, obtain the target text based on the processed initial text, and upload the target text to the blockchain;

If not, the action data of the smart cursor is determined again according to the real-time status information.
A grammar error correction device, which includes:

The preprocessing module is used to obtain the initial text, and insert a smart cursor that can perform actions at a preset position of the initial text;

The state acquisition module is used to mark the real-time state of the initial text with the smart cursor to obtain real-time state information;

An action determining module, configured to determine the action data of the smart cursor according to the real-time status information;

The action execution module is configured to use the smart cursor to process the initial text based on the action data to obtain the target text.
A computer device, wherein the computer device includes a memory, a processor, and a computer program that is stored in the memory and can run on the processor, and the processor implements the grammatical error correction method when the computer program is executed. The following steps:

Acquiring the initial text, and inserting a smart cursor that can perform actions at a preset position of the initial text;

Mark the initial text with the smart cursor in real-time status to obtain real-time status information;

Using an error correction model to determine the action data of the smart cursor according to the real-time status information;

Using the smart cursor to process the initial text based on the action data to obtain the target text.
9. The computer device according to claim 9, wherein the real-time status marking of the initial text with the smart cursor to obtain real-time status information comprises the following:

Serialize the initial text to obtain first processed data;

The smart cursor is located in real time, real-time position information of the smart cursor is obtained, the first processed data is marked based on the position information, and the first processed data with the real-time position mark of the cursor is obtained as real-time status information.
9. The computer device according to claim 9, wherein the determining the action data of the smart cursor by using an error correction model according to the real-time status information comprises the following:

Use a neural network to perform mapping processing on the real-time status information to obtain the first data;

Performing element averaging processing on the first data to obtain second data;

A deep reinforcement learning model is used to process the second data to determine the smart cursor action data.
The computer device according to claim 9, wherein, before determining the action data of the smart cursor according to the real-time status information, training the error correction model includes the following:

Obtaining training samples, processing the training samples by using a neural network, and obtaining sample processing data;

Use the action network and state network in the deep reinforcement learning model to process the training sample processing data to obtain the initial action strategy and value function;

Using loss function processing to obtain sample action data based on the initial action strategy and value function;

Use a compiler to compile the sample action data, and obtain reward and punishment data according to the compiling result;

The loss function and parameters in the error correction model are adjusted based on the reward and punishment data, and processed again until the training process is completed, and a trained error correction model is obtained.
9. The computer device according to claim 9, wherein said using said smart cursor to process said initial text based on said action data comprises the following steps:

Obtaining a corresponding data type based on the action data;

When the data type is edit data, edit the initial text according to the action data, and update the real-time status information based on the position information of the smart cursor;

When the data type is navigation data, move the smart cursor according to the action data and update the real-time status information based on the position information of the smart cursor.
9. The computer device according to claim 9, wherein after the initial text is processed and before the target text is obtained, the method further comprises the following steps:

Acquiring current position information of the smart cursor, and judging whether the smart cursor is at the end of the initial text based on the position information;

If yes, obtain the target text based on the processed initial text, and upload the target text to the blockchain;

If not, the action data of the smart cursor is determined again according to the real-time status information.
A computer-readable storage medium includes multiple storage media, and each storage medium stores a computer program, wherein the computer programs stored in the multiple storage media jointly implement the grammatical error correction when executed by a processor The following steps of the method:

Acquiring the initial text, and inserting a smart cursor that can perform actions at a preset position of the initial text;

Mark the initial text with the smart cursor in real-time status to obtain real-time status information;

Using an error correction model to determine the action data of the smart cursor according to the real-time status information;

Using the smart cursor to process the initial text based on the action data to obtain the target text.
The computer-readable storage medium according to claim 15, wherein the real-time status marking of the initial text with the smart cursor to obtain real-time status information comprises the following:

Serialize the initial text to obtain first processed data;

The smart cursor is located in real time, real-time position information of the smart cursor is obtained, the first processed data is marked based on the position information, and the first processed data with the real-time position mark of the cursor is obtained as real-time status information.
15. The computer-readable storage medium according to claim 15, wherein the determining the action data of the smart cursor according to the real-time status information using an error correction model comprises the following:

Use a neural network to perform mapping processing on the real-time status information to obtain the first data;

Performing element averaging processing on the first data to obtain second data;

A deep reinforcement learning model is used to process the second data to determine the smart cursor action data.
The computer-readable storage medium according to claim 15, wherein, before determining the action data of the smart cursor according to the real-time status information, training the error correction model includes the following:

Obtaining training samples, processing the training samples by using a neural network, and obtaining sample processing data;

Use the action network and state network in the deep reinforcement learning model to process the training sample processing data to obtain the initial action strategy and value function;

Using loss function processing to obtain sample action data based on the initial action strategy and value function;

Use a compiler to compile the sample action data, and obtain reward and punishment data according to the compiling result;

The loss function and parameters in the error correction model are adjusted based on the reward and punishment data, and processed again until the training process is completed, and the trained error correction model is obtained.
15. The computer-readable storage medium according to claim 15, wherein said using said smart cursor to process said initial text based on said action data comprises the following steps:

Obtaining a corresponding data type based on the action data;

When the data type is edit data, edit the initial text according to the action data, and update the real-time status information based on the position information of the smart cursor;

When the data type is navigation data, move the smart cursor according to the action data and update the real-time status information based on the position information of the smart cursor.
15. The computer-readable storage medium according to claim 15, wherein after the initial text is processed and before the target text is obtained, the method further comprises the following steps:

Acquiring current position information of the smart cursor, and judging whether the smart cursor is at the end of the initial text based on the position information;

If yes, obtain the target text based on the processed initial text, and upload the target text to the blockchain;

If not, the action data of the smart cursor is determined again according to the real-time status information.