Text proofreading method and device for man-machine cooperation feedback learning

Abstract

The invention relates to a text proofreading method and device for man-machine cooperation feedback learning, and the method comprises the steps: carrying out the preprocessing of a to-be-proofreading text, and generating a character sequence of the to-be-proofreading text; performing machine proofreading on the character sequence for multiple times, and updating the character sequence in sequence in response to a corresponding manual confirmation result after each time of machine proofreading; and in response to a manual proofreading result of the character sequence after machine proofreading, generating a final proofreading result and outputting a proofreading text. By combining machine proofreading and manual confirmation, efficient and accurate text proofreading is realized. Through multi-round feedback learning and optimization of a training word library, the intelligence degree and adaptability of the proofreading system are remarkably improved, and the text proofreading requirements in different scenes are met.

Description

Technical Field

[0001] The present invention relates to the technical field of text proofreading, and in particular to a text proofreading method and device for human-computer collaborative feedback learning. Background Art

[0002] In the field of text proofreading, traditional proofreading methods mainly rely on manual work, which is not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in unstable proofreading quality. With the development of natural language processing and machine learning technologies, automated methods based on machine proofreading have gradually become a research hotspot. However, existing machine proofreading technologies still have problems with insufficient accuracy and robustness, especially when dealing with complex grammatical structures and polysemous words. In addition, relying entirely on machines for proofreading is difficult to adapt to the diverse needs in different scenarios.

[0003] In order to solve the above problems, the human-machine collaborative text proofreading method came into being. By combining machine proofreading and manual proofreading, the advantages of machine fast processing and human semantic understanding can be fully utilized to significantly improve proofreading efficiency and accuracy.

[0004] However, current human-machine collaborative proofreading methods generally lack an effective feedback learning mechanism and are unable to fully utilize the data generated by each proofreading for model optimization and self-improvement. Summary of the invention

[0005] The purpose of the present invention is to provide a text proofreading method and device for human-computer collaborative feedback learning to solve the problem that the prior art lacks an effective feedback learning mechanism.

[0006] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention provides a text proofreading method of human-machine collaborative feedback learning, the method comprising:

[0007] Preprocess the text to be proofread to generate a character sequence of the text to be proofread;

[0008] Performing multiple machine proofreadings on the character sequence, and updating the character sequence in sequence in response to a manual confirmation result corresponding to each machine proofreading;

[0009] In response to the manual proofreading result of the machine-proofread character sequence, a final proofreading result is generated and a proofreading text is output.

[0010] As a further improvement of an embodiment of the present invention, the method further includes that the preprocessing includes cutting the text to be processed into paragraphs, sentences, words and punctuation characters to obtain a character sequence containing text position offset information.

[0011] As a further improvement of an embodiment of the present invention, the method further includes that the performing multiple machine proofreadings on the character sequence includes:

[0012] The character sequence is machine proofread using a pre-trained error classification feature model, and a preliminary proofreading result is generated by detecting potential errors in the character sequence.

[0013] As a further improvement of an embodiment of the present invention, the method further includes that the “updating the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading” includes:

[0014] Manually confirm the preliminary proofreading result, update the character sequence, and perform machine proofreading again to determine whether new potential errors appear in the updated character sequence;

[0015] If yes, repeat the machine proofreading and manual confirmation operations and update the character queue in sequence;

[0016] If not, the character sequence after machine proofreading is output.

[0017] As a further improvement of an embodiment of the present invention, the method further includes constructing a training word library based on the results of multiple machine proofreadings and the results of manual proofreadings;

[0018] Training the error classification feature model based on the training vocabulary;

[0019] The error classification feature model includes an error detection model and an error learning model;

[0020] The error detection model includes error detection models for homophone error detection, similar character error detection, detection of correspondence between traditional and simplified Chinese characters, and detection of fixed expressions of time and date; the error learning model includes a BERT deep learning model.

[0021] As a further improvement of an embodiment of the present invention, the method further includes: the training step of the error detection model includes:

[0022] Construct multi-branch error detection rules that include homophones, similar characters, correspondence between traditional and simplified characters, and fixed expressions of time and date;

[0023] The text to be proofread is detected based on the multi-branch error detection rules. When the text to be proofread meets any error detection rule, it is confirmed that it belongs to the corresponding error type and is stored in the training word library.

[0024] As a further improvement of an embodiment of the present invention, the method further includes that the training step of the error learning model includes:

[0025] Constructing positive and negative samples based on the training word library, and generating sample sequences by segmenting the positive and negative samples; the positive and negative samples include text data marked as error words and non-error words;

[0026] Marking the positions of the error words in the sample sequence to generate a training sequence;

[0027] Using the BERT deep learning model to perform classification training on the training sequence to obtain a classification feature model;

[0028] The text to be proofread is input into the classification feature model to generate a corresponding matching score, and a classification prediction result is output for each character sequence position to determine the potential error type.

[0029] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention provides a text proofreading device for human-machine collaborative feedback learning, wherein the system comprises a text cutting module, a machine proofreading module and a proofreading feedback module;

[0030] The text cutting module is used to pre-process the text to be proofread and generate a character sequence of the text to be proofread;

[0031] The machine proofreading module is used to perform multiple machine proofreadings on the character sequence, and update the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading;

[0032] The proofreading feedback module is used to generate a final proofreading result and output a proofreading text in response to the manual proofreading result of the character sequence after the machine proofreading is completed.

[0033] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention further provides an electronic device, comprising a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the text proofreading method of human-computer collaborative feedback learning as described above are implemented.

[0034] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention further provides a storage medium, wherein the storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, the steps in the text proofreading method of human-computer collaborative feedback learning as described above are implemented.

[0035] Compared with the prior art, the present invention provides a text proofreading method and device for human-machine collaborative feedback learning, which realizes efficient and accurate text proofreading by combining machine proofreading and manual confirmation. Through multiple rounds of feedback learning and optimization of the training vocabulary, the intelligence and adaptability of the proofreading system are significantly improved, meeting the text proofreading needs in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is an overall flow chart of the text proofreading method of human-machine collaborative feedback learning described in the present invention.

[0037] Figure 2 It is a schematic diagram of the structure of the text proofreading device for human-machine collaborative feedback learning described in the present invention. DETAILED DESCRIPTION

[0038] The present invention will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by a person skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0039] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0040] In the first embodiment of the present invention, the present invention provides a text proofreading method for human-machine collaborative feedback learning, such as Figure 1 As shown, the method includes,

[0041] S1: preprocess the text to be proofread to generate a character sequence of the text to be proofread;

[0042] S2: performing machine proofreading on the character sequence for multiple times, and updating the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading;

[0043] S3: In response to the manual proofreading result of the machine-proofread character sequence, generate a final proofreading result and output a proofreading text.

[0044] In a specific embodiment of the present invention, the preprocessing includes cutting the text to be processed into paragraphs, sentences, words and punctuation characters to obtain a character sequence containing text position offset information.

[0045] Furthermore, after the information to be processed is preprocessed, the first version of the text to be proofread is obtained A=(A1, A2, A3, ..., A n ), where n is the number of characters contained in the text.

[0046] In a specific embodiment of the present invention, the character sequence is subjected to multiple machine proofreadings, specifically:

[0047] The character sequence is machine proofread using a pre-trained error classification feature model, and a preliminary proofreading result is generated by detecting potential errors in the character sequence.

[0048] Furthermore, the text to be proofread is proofread through a first round of machine proofreading and error correction to obtain first round proofreading error correction information. The error correction information includes zero or more position offsets of text error information, error correction suggestions, and other information.

[0049] For example, the sentence "The cultural and sports center is located at the intersection of Zhongshan Road and Yanjiang Road, and will meet the cultural and sports needs of the surrounding residents" is used as an example. After proofreading by the machine model, it is recognized that the word "intersection" is wrong. The machine proofreading result R = (R1, R2, ..., R i ), where i is the number of machine proofreading marks, and Ri is the proofreading item information tuple. In this example, R1 = (15, 16, 0, "error word", "intersection", ...), where machine proofreading 15, 16 is the position offset of the suspected error character in the text. In item 3, 0 means that the current proofreading result is not confirmed, and 1 means that the proofreading item has been confirmed. "Wrong word" is the error type, and "intersection" is the correction word. In addition, it may include but is not limited to correction suggestions, paragraph numbers, etc.

[0050] In a specific embodiment of the present invention, in response to the manual confirmation result corresponding to each machine proofreading, the character sequence is updated in sequence, specifically,

[0051] Manually confirm the preliminary proofreading result, update the character sequence, and perform machine proofreading again to determine whether new potential errors appear in the updated character sequence;

[0052] If yes, repeat the machine proofreading and manual confirmation operations and update the character queue in sequence;

[0053] If not, the character sequence after machine proofreading is output.

[0054] Furthermore, in the second round of proofreading, the proofreader confirms the results of the first round of proofreading, that is, the "intersection" in the example should be changed to "intersection", and based on the proofreading result R, it is updated to obtain S = (S1, S2, ..., S i ). For example, S1 = (15, 16, 1, "error word", "intersection", ...). At this time, "intersection" in the text has been adjusted to "intersection", and the new text is "The Cultural and Sports Center is located at the intersection of Zhongshan Road and Yanjiang Road, and will meet the various cultural and sports needs of residents on the border of the state." The second version text B = (B1, B2, B3, ..., B m ), where m is the number of characters in the text, and m and n may be different.

[0055] Furthermore, in the third round of proofreading, the machine proofreading model proofreads the modified text B and obtains the machine proofreading result W = (W1, W2, ..., W j), where j is the number of machine proofreading marks. For example, W1 = (22, 23, 0, "suspicious item", "nearby",...). The proofreader confirms the proofreading result W and does not adopt this suggestion, and retains the result in X = (X1, X2,..., X j ), where X1 = (22, 23, 0, "suspicious item", "nearby",...).

[0056] Furthermore, the information processor checks the text to be proofread, marks new errors, and saves the proofreading result as Y = (Y1, Y2,..., Y k ). For example, Y1 = (20, 21, 0, "error word", "meet",...), and there is a grammar collocation error in "bear... needs", which is modified to "meet... needs".

[0057] Even further, the proofreading result Y is reconfirmed by the information processor, and the proofreading result meta-items are updated to obtain the proofreading result Z = (Z1, Z2,..., Z k ). For example, Z1 = (20, 21, 1, "error word", "meet",...), where the value 1 indicates that this proofreading suggestion is adopted. The information processor modifies the text to be proofread accordingly to obtain the third version of the text C = (C1, C2,..., C o ), where o is the number of characters contained in the text, and o may be different from m and n.

[0058] Even further, 3 versions of the text A, B, C and 6 proofreading version results R, S, W, X, Y, Z are obtained through the above steps. The relevant data is automatically synchronized to the training word library during the above process.

[0059]

[0060] It should be noted that based on the fact that each version of the data has been stored in the training word library, each version of the text and each proofreading version can be cross-compared with each other to form multiple groups of sentence pairs. The sentence pairs contain information such as the position information of the error words in the previous round, the confirmation of the proofreading result, the proofreading suggestion, and the sentence context. One combination method is to combine 1 item of text with 2 items of proofreading results. For example, the sentence pair E ARS = (A, R, S) is combined by the proofreading text A and the proofreading results R and S, and it can be confirmed that there is an error word "confluence" in A, and the error marked by the proofreading result R is adopted. Similarly, the sentence pair E BWY = (B, W, X) can confirm the machine proofreading result W. Another combination method is to combine 2 items of text and 1 item of result. For example, the sentence pair E ABY = (A, B, Y) is combined by the proofreading texts A and B and the proofreading result Y. Similarly, there is E ACY=(A, C, Y). In this way, unrecognized error words can be identified for the proofreading text A, and the information of manual annotation items can be strengthened.

[0061] In a specific embodiment of the present invention, a training word bank is constructed based on the results of multiple machine proofreading and the results of manual proofreading;

[0062] The error classification feature model is trained based on the training word bank;

[0063] The error classification feature model includes an error detection model and an error learning model;

[0064] The error detection model includes error detection models for homophone error detection, shape-similar character error detection, traditional Chinese and simplified Chinese character correspondence detection, and fixed expression detection of time and date; the error learning model includes a BERT deep learning model.

[0065] Furthermore, the training steps of the error detection model are specifically as follows:

[0066] Construct a multi-branch error detection rule including homophones, shape-similar characters, traditional Chinese and simplified Chinese character correspondence, and fixed expression of time and date;

[0067] Based on the multi-branch error detection rule, the text to be proofread is detected. When the text to be proofread meets any error detection rule, it is confirmed that it belongs to the corresponding error type and is stored in the training word bank.

[0068] It should be noted that the error detection rules include: (1) Based on homophone errors: for example, in "when it is raining heavily", "侯" should be "候" because "时侯" and "时候" have the same pronunciation but different usages. (2) Shape-similar character errors: such as "the water in the river is very clear撤", where "撤" should be changed to "澈" because "撤" and "澈" are similar in shape but different in meaning. (3) Traditional Chinese and simplified Chinese character correspondence check. (4) Fixed expression error detection of time and date, including regular expression matching and specific format detection, such as "year-month-day hour-minute".

[0069] Furthermore, the training steps of the error learning model are specifically as follows:

[0070] Positive and negative samples are constructed based on the training word bank, and a sample sequence is generated by splitting the positive and negative samples; the positive and negative samples include text data labeled as error words and non-error words;

[0071] The positions of error words in the sample sequence are labeled to generate a training sequence;

[0072] The BERT deep learning model is used to classify and train the training sequence to obtain a classification feature model;

[0073] The text to be proofread is input into the classification feature model to generate a corresponding matching score, and a classification prediction result is output for each character sequence position to determine the potential error type.

[0074] It should be noted that the error learning model is constructed by combining sequence annotation with deep learning, including: (1) merging the positive and negative samples of the first-level classification into a corpus with a large amount of data. (2) Sentences in the classification samples are cut at the word level to obtain character sequences. (3) Sequence annotation is performed on the positions of the error words in the sequence to obtain the BERT deep learning model training sequence = [CLS, 0, 0.. 111., 0], where CLS corresponds to the first-level classification label, the value 0 corresponds to the non-error text position in the sentence, and the value 1 corresponds to the error word annotation position. (4) The BERT model is used to classify the word library sequence to obtain a classification feature model. (5) The sentence to be classified is input into the feature model to obtain multiple matching degrees. The one with the highest ranking among the multiple matching degrees or the classification prediction score higher than the preset is determined as the classification.

[0075] It should be noted that the framework of primary and secondary classification is divided from different dimensions based on the actual business situation and expert opinions. For example, the primary classification is divided into (but not limited to): text punctuation errors, knowledge errors, and content risks. Among them, text punctuation errors focus on grammatical level norms, knowledge errors focus on the level of fixed knowledge expression, and content risks focus on semantic level content security analysis.

[0076] The construction of classification samples includes: (1) refining the secondary classification based on the primary classification, and building a preset word library based on the secondary classification. The word library includes an error word dictionary and an error example sentence set, and each sentence in the example sentence set includes 0, 1 or more error words and corresponding error correction words (including error correction suggestions); (2) extracting samples from the secondary preset word library in proportion to form a primary classification word library sample, which includes an error word word library and a text sentence set.

[0077] Furthermore, after the proofreading sample is classified into the first level through the error classification feature model, the second level classification is performed to obtain the predicted value of the second level classification. When the predicted value meets the set threshold, the classification vocabulary detection mode is obtained and the vocabulary conformity detection is performed. The implementation method is as follows:

[0078] In one embodiment, a mathematical statistical method is used. The number of times the wrong word and its preceding word appear in the corpus, and the number of times the wrong word and its succeeding word appear in the corpus are counted. The above-mentioned preset corpus can be generated based on a large corpus, which is not specifically limited in the present disclosure. When the number of occurrences exceeds a preset threshold, the entry is synchronized to the training lexicon in combination with the proofreading result.

[0079] In one embodiment, the edit distance is calculated by using positive and negative samples. The edit distance is a method to measure the difference between two strings, which is the minimum number of single-character editing operations required to convert one string into another. By comparing the annotated words in the erroneous sentence with those in the correct sentence. When the edit distance meets the preset threshold, the entry is synchronized to the training word library in combination with the proofreading result.

[0080] In one embodiment, the sentence pairs of positive and negative samples are vectorized by BERT to obtain the corresponding vector matrix. Taking EABY = (A, B, Y) as an example, MAT is obtained by BERT vectorization. A , MAT B , and obtain the rationality score of each character in A and B. MAT A , MAT B The proofreading result Y is input into the transformation function to obtain the vectorized G containing the proofreading suggestion. A , G B Calculate G A , G B The sum of the vectors in is calculated, and then the difference between the two is obtained to obtain the semantic features of the marked word error word. When the feature meets the threshold, the marked words and recognition rules in the proofreading are stored in the training word library.

[0081] In the second embodiment of the present invention, the present invention provides a text proofreading device for human-machine collaborative feedback learning, comprising a text cutting module 1, a machine proofreading module 2 and a proofreading feedback module 3;

[0082] The text cutting module 1 is used to pre-process the text to be proofread and generate a character sequence of the text to be proofread;

[0083] The machine proofreading module 2 is used to perform multiple machine proofreadings on the character sequence, and update the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading;

[0084] The proofreading feedback module 3 is used to generate a final proofreading result and output a proofreading text in response to the manual proofreading result of the character sequence after the machine proofreading is completed.

[0085] In a third embodiment of the present invention, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the text proofreading method of human-computer collaborative feedback learning as described above are implemented.

[0086] In a fourth embodiment of the present invention, the present invention provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps in the text proofreading method of human-computer collaborative feedback learning as described above.

[0087] In summary, the text proofreading method and device of human-machine collaborative feedback learning provided by the present invention realizes efficient and accurate text proofreading by combining machine proofreading and manual confirmation. Through multiple rounds of feedback learning and optimization of the training vocabulary, the intelligence and adaptability of the proofreading system are significantly improved, meeting the text proofreading needs in different scenarios.

[0088] It should be understood that although this specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each implementation mode may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0089] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the modules described above can refer to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0090] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present implementation scheme.

[0091] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0092] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for a computer system (which can be a personal computer, a server, or a network system, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0093] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned implementation modes, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned implementation modes, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various implementation modes of the present application.

Claims

1. A text proofreading method based on human-computer collaborative feedback learning, characterized in that: include, Preprocess the text to be proofread to generate a character sequence of the text to be proofread; Performing multiple machine proofreadings on the character sequence, and updating the character sequence in sequence in response to a manual confirmation result corresponding to each machine proofreading; In response to the manual proofreading result of the machine-proofread character sequence, a final proofreading result is generated and a proofreading text is output.

2. The text proofreading method of human-computer collaborative feedback learning according to claim 1, characterized in that: The preprocessing includes cutting the text to be processed into paragraphs, sentences, words and punctuation characters to obtain a character sequence containing text position offset information.

3. The text proofreading method of human-machine collaborative feedback learning according to claim 2, characterized in that: The multiple machine proofreading of the character sequence comprises: The character sequence is machine proofread using a pre-trained error classification feature model, and a preliminary proofreading result is generated by detecting potential errors in the character sequence.

4. The text proofreading method of human-machine collaborative feedback learning according to claim 3 is characterized by: The "updating the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading" includes: Manually confirm the preliminary proofreading result, update the character sequence, and perform machine proofreading again to determine whether new potential errors appear in the updated character sequence; If yes, repeat the machine proofreading and manual confirmation operations and update the character queue in sequence; If not, the character sequence after machine proofreading is output.

5. The text proofreading method of human-machine collaborative feedback learning according to claim 1, characterized in that: Also includes, Build a training vocabulary based on multiple machine proofreading results and manual proofreading results; Training the error classification feature model based on the training vocabulary; The error classification feature model includes an error detection model and an error learning model; The error detection model includes error detection models for homophone error detection, similar character error detection, detection of correspondence between traditional and simplified Chinese characters, and detection of fixed expressions of time and date; the error learning model includes a BERT deep learning model.

6. The text proofreading method of human-machine collaborative feedback learning according to claim 5, characterized in that: The training step of the error detection model includes: Construct multi-branch error detection rules that include homophones, similar characters, correspondence between traditional and simplified characters, and fixed expressions of time and date; The text to be proofread is detected based on the multi-branch error detection rules. When the text to be proofread meets any error detection rule, it is confirmed that it belongs to the corresponding error type and is stored in the training word library.

7. The text proofreading method of human-machine collaborative feedback learning according to claim 6, characterized in that: The training steps of the error learning model include: Constructing positive and negative samples based on the training word library, and generating sample sequences by segmenting the positive and negative samples; the positive and negative samples include text data marked as error words and non-error words; Marking the positions of the error words in the sample sequence to generate a training sequence; Using the BERT deep learning model to perform classification training on the training sequence to obtain a classification feature model; The text to be proofread is input into the classification feature model to generate a corresponding matching score, and a classification prediction result is output for each character sequence position to determine the potential error type.

8. A text proofreading device for human-machine collaborative feedback learning, characterized in that: Including text cutting module, machine proofreading module and proofreading feedback module; The text cutting module is used to pre-process the text to be proofread and generate a character sequence of the text to be proofread; The machine proofreading module is used to perform multiple machine proofreadings on the character sequence, and update the character sequence in sequence in response to the manual confirmation result corresponding to each machine proofreading; The proofreading feedback module is used to generate a final proofreading result and output a proofreading text in response to the manual proofreading result of the character sequence after the machine proofreading is completed.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the text proofreading method for human-computer collaborative feedback learning as described in any one of claims 1-7 are implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps in the text proofreading method of human-computer collaborative feedback learning as described in any one of claims 1 to 7 are implemented.

Images