Foreign affair table intelligent filling error correction method and system based on natural language processing

Abstract

The invention relates to the technical field of data processing, in particular to an intelligent foreign form filling and error correction method and system based on natural language processing. The method comprises the following steps: acquiring text images of a historical corpus and vocabularies; screening segmented words to be corrected; determining the ambiguity of the module based on the frequency proportion of the input vocabulary and the vocabulary type; determining a first possibility based on the co-occurrence probability of the adjacent vocabularies of the to-be-corrected segmented word and the similar segmented word and the similarity of the to-be-corrected segmented word and the similar segmented word; determining a second possibility through an error correction result corresponding to the similar word segmentation history and the image similarity of the word segmentation to be corrected and the similar word segmentation; and determining the total possibility and the necessity of word segmentation error correction through the first possibility, the second possibility and the ambiguity, and completing word segmentation error correction. The error correction accuracy is improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to an intelligent form-filling and error-correcting method and system for foreign affairs forms based on natural language processing. Background Art

[0002] In recent years, natural language processing (NLP) technology has made remarkable progress in information extraction, text understanding, semantic analysis, etc., providing strong technical support for intelligent processing of unstructured or semi-structured texts. Applying advanced natural language processing technology to the intelligent form-filling and error-correcting of foreign affairs forms has important practical significance and application value. By constructing an intelligent form-filling and error-correcting system for foreign affairs forms based on NLP, it is possible to achieve real-time semantic understanding of the content input by users, automatic extraction of key information, intelligent verification of filling specifications, as well as error prompts and correction suggestions, thereby effectively improving the filling accuracy of foreign affairs forms, reducing the pressure of manual review, shortening the approval cycle, and enhancing the intelligent, standardized, and efficient level of foreign affairs management work.

[0003] During the process of filling intelligent forms, when OCR recognizes paper forms with poor handwriting or printing quality, it is easy to misrecognize some Chinese characters (such as misrecognizing "张" as "弓长", "婚" as "昏"), especially for connected or scribbled handwriting. Simply relying on edit distance to match the corpus cannot consider the context semantics; it ignores the actual writing image features (i.e., visual similarity); resulting in poor error-correcting effects. Summary of the Invention

[0004] To solve the technical problem of poor vocabulary error-correcting effect, this application provides an intelligent form-filling and error-correcting method and system for foreign affairs forms based on natural language processing. The specific technical solutions adopted are as follows: In the first aspect, this application proposes an intelligent form-filling and error-correcting method for foreign affairs forms based on natural language processing. The method includes the following steps: Obtain the corpus corresponding to each module of the foreign affairs form and the text image of the vocabulary; After inputting the text, perform word segmentation on the text to obtain input word segments, and screen the word segments to be error-corrected from the input word segments through the corpus; for each module, count the types of input word segments, and determine the ambiguity of each module based on the frequency of the input word segments, the average frequency of all input word segments, and the types of input word segments; Determine similar word segments based on the edit distance between the word segments to be error-corrected and the word segments in the corpus; calculate the co-occurrence probability between the adjacent word segments of the word segments to be error-corrected and the similar word segments, and determine the first possibility of each similar word segment being the word segment to be error-corrected based on the co-occurrence probability and the edit distance between the word segments to be error-corrected and the similar word segments; The error correction frequency of the word to be corrected is determined based on the frequency of word segments that were corrected to similar words in history; the second possibility of similar word segments is obtained based on the error correction frequency of the word to be corrected and the similarity between the word to be corrected and the text image of similar words. The total probability is determined based on the first and second probabilities of the error-correcting word segmentation; the necessity of word segmentation error correction is determined based on the total probability of similar word segmentation and the ambiguity of the module; and word segmentation error correction is performed based on the necessity of word segmentation error correction and the total probability.

[0005] In the aforementioned scheme, this application significantly improves the accuracy of OCR recognition and correction of errors (especially similar-looking characters and blurred handwritten characters) by integrating multi-dimensional analysis of text edit distance, contextual co-occurrence probability, image morphological similarity, and historical error correction records. An image-level verification mechanism is introduced, using edge detection and shape context algorithms to compare the historical misidentified images of the word to be corrected with candidate words, improving the reliability of the error correction basis from the perspective of physical writing morphology and avoiding miscorrections caused by pure text matching. By calculating the ambiguity of the content in each filling module, the diversity and openness of the content in that module are dynamically evaluated. Strong error correction is applied to modules with fixed content, while the error correction intensity is reduced in modules with open content, avoiding overcorrection of reasonable new words.

[0006] In one embodiment, the frequency of the input segment is the number of times the input segment appears in history divided by the total number of times all input segments appear.

[0007] In one embodiment, the ambiguity is positively correlated with the difference between the frequency of each input word segment and the average frequency of all input words, as well as the number of types of input words.

[0008] In one embodiment, the first probability is positively correlated with the first product and the second product, respectively, and negatively correlated with the edit distance between the word segment to be corrected and the similar word segment; the first product and the second product are the co-occurrence probabilities of the adjacent word segment and the similar word segment of the word segment to be corrected, respectively.

[0009] In one embodiment, the method for determining the correction frequency of the word to be corrected based on the frequency of words that were corrected to similar words in history is as follows: For each similar word segment, retrieve the word segment that was corrected to that similar word segment in the history and record it as a corrected word segment. Calculate the correction frequency of each corrected word segment, where the correction frequency is the ratio of the number of each corrected word segment to the total number of corrected words. If a corrected word segment includes a word to be corrected, then the correction frequency of the corrected word segment is the same as the correction frequency of the word to be corrected; otherwise, the correction frequency of the word to be corrected is 0.

[0010] In one embodiment, the second probability is positively correlated with text image similarity and error correction frequency, respectively.

[0011] In one embodiment, the total probability is positively correlated with the first probability and the second probability, respectively.

[0012] In one embodiment, the necessity of word segmentation correction is positively correlated with the total probability and negatively correlated with the ambiguity.

[0013] In one embodiment, the method for correcting word segmentation based on the necessity and total probability of word segmentation correction is as follows: The necessity of word segmentation correction is compared with the correction threshold. If the necessity of word segmentation correction is less than the correction threshold, the word segment to be corrected will not be corrected. If the necessity of word segmentation correction is greater than the correction threshold, then correction is required, and the most likely similar word segment is taken as the corrected word segment.

[0014] On the other hand, this application also provides an intelligent form filling and error correction system for foreign affairs forms based on natural language processing, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described intelligent form filling and error correction methods for foreign affairs forms based on natural language processing.

[0015] The beneficial effects of this application are as follows: This application significantly improves the accuracy of OCR recognition and correction of errors (especially similar-looking characters and blurred handwritten characters) by integrating multi-dimensional analysis of text edit distance, contextual co-occurrence probability, image morphological similarity, and historical error correction records. It introduces an image-level verification mechanism, using edge detection and shape context algorithms to compare the word to be corrected with historical misidentified images of candidate words, enhancing the reliability of the error correction basis from the perspective of physical writing morphology and avoiding false corrections caused by pure text matching. By calculating the ambiguity of the content in each filling module, the diversity and openness of the content in that module are dynamically evaluated. Strong error correction is applied to modules with fixed content, while the error correction intensity is reduced in modules with open content, avoiding over-correction of reasonable new words. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for intelligent filling and error correction of foreign affairs forms based on natural language processing, provided as an embodiment of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent form filling and error correction method and system for foreign affairs forms based on natural language processing proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] An Example of an Intelligent Form Filling and Error Correction Method and System Based on Natural Language Processing for Foreign Affairs Forms: The following, in conjunction with the accompanying drawings, details the specific scheme of the intelligent form filling and error correction method and system for foreign affairs forms based on natural language processing provided in this application.

[0021] Please see Figure 1 The diagram illustrates a flowchart of an intelligent form filling and error correction method for foreign affairs forms based on natural language processing, according to an embodiment of this application. The method includes the following steps: Step S001: Obtain text images of historical corpora and vocabulary.

[0022] This solution utilizes existing OCR equipment to recognize paper forms, obtain user input, and read user input modules. For foreign affairs forms, OCR technology is used to recognize all characters in the form content. Descriptive fields such as "name" and "gender" are considered individual modules, and character matching is primarily used to recognize each module. Additionally, information entered into the form, such as "Zhang San" or "male," is considered user input. The main objective of this solution is to intelligently recognize and correct erroneous user input information and fill it into the corresponding modules.

[0023] For each module of the target foreign affairs form, the correct sample data corresponding to each module is pre-acquired, i.e., the historical correct vocabulary set of the module content is obtained, and the vocabulary set corresponding to each module is recorded as the corpus corresponding to that module. That is, each type of form corresponds to a set of corpora, and the corpus and the module of the form are in one-to-one correspondence. The text image of each word in the corpus is saved for subsequent determination of the similarity between the text to be corrected and the word segmentation of the corpus based on the image.

[0024] At this point, the corpus and text images corresponding to each module have been obtained.

[0025] Step S002: Filter the word segments to be corrected; determine the fuzziness of the module based on the frequency ratio of the input words and the word categories.

[0026] For the user's input text, the text is first segmented into words, and then each word is checked to see if it exists in the corpus. If the word is in the corpus to which the module belongs, it is considered a normal word; otherwise, it is considered an abnormal word and needs to be corrected. This segmented word is then recorded as a word to be corrected. In this embodiment, the existing jieba segmentation algorithm is used to segment the input text. The resulting word segments for each module in the table are then obtained.

[0027] In the process of text correction, there are three main types of errors: word-level errors, grammatical errors, and semantic errors. This application mainly considers the situation where some words are not accurately recognized when converting paper content to electronic spreadsheets in the process of OCR recognition. Therefore, this application mainly corrects word-level errors and does not include the correction of non-text information such as ID cards and phone numbers.

[0028] Considering that the number of entries varies across different areas of the table, and the standards for filling in the entries also differ, some modules have relatively fixed content to fill in, while others do not. Therefore, in the process of correcting abnormal words, it is also necessary to determine whether the word segmentation is a reasonable new word segmentation that may have been added to the content of that module, rather than a word segmentation that was actually erroneous during the acquisition process. Therefore, this application analyzes the word segmentation content of each module, calculates the ambiguity of the content of each module, analyzes the ambiguity of the module's identification and correction, and thus determines whether to correct abnormal word segmentation.

[0029] Specifically, for each module, the input text is segmented using a word segmentation method, and the resulting words are recorded as input words. The frequency of each input word in the historical content is counted, and the number of word categories is also counted. In this embodiment, the number of gender categories is 2, marital status is 2, and education level is 6. The content richness of the module can then be obtained. The higher the content richness of the module, the greater the probability of other words appearing during recognition, and vice versa.

[0030] The ambiguity is determined based on the difference between the frequency of each input word segment and the average frequency of all input words, as well as the number of types of input words.

[0031] The ambiguity is positively correlated with the difference between the frequency of each input word segment and the average frequency of all input words, as well as the number of types of input words.

[0032] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large. The specific relationship is determined by the actual application, and this application does not impose any special restrictions.

[0033] The preferred expression for ambiguity is: , This indicates the frequency of the i-th input word segmentation. This represents the average frequency of all input words. Indicates the type of input word segmentation. Represents a linear normalization function. This indicates the ambiguity of the content entered in the Lth module.

[0034] The more types of input word segments appearing in the module, and the greater the difference in the frequency of the input word segments, the higher the ambiguity of the module's content. That is, the greater the possibility that the types of input word segments appearing in the module exceed the word segments outside the word segmentation samples.

[0035] At this point, the fuzziness of the content entered in each module has been obtained.

[0036] Step S003: Determine the first possibility based on the co-occurrence probability of adjacent words and similar words of the word to be corrected and the similarity between the word to be corrected and the similar words.

[0037] In existing text segmentation error correction processes, most methods use edit distance to calculate the target segmentation and select the segmentation with the closest edit distance from the corpus as the corrected segmentation. However, this may not reflect handwriting characteristics; for example, handwritten and computer-displayed text are easily identified as corresponding incorrect segmentations. Therefore, this application also incorporates analysis of the target module's historical error correction records and the target segmentation error correction frequency to refine the obtained edit distance and increase the reliability of the segmentation error correction.

[0038] The process of error correction based on edit distance obviously does not take into account the context, which may lead to correction errors. Therefore, this application combines the context and the ambiguity of the content module to correct word segmentation.

[0039] Specifically, obtain the adjacent tokens before and after the token to be corrected. The adjacent tokens are obtained according to the positions of the token division results in the original samples for each sample data. At the same time, the tokens before and after the abnormal word in the original recognized text after token division are respectively recorded as the previous token and the subsequent token. In this embodiment, there is a recognized content, "my / English / is very beautiful", where "English" is the token to be corrected, "of" is the previous token, and "is very beautiful" is the subsequent token. Then, it can be corrected by calculating the probabilities and token similarities of the corpus tokens and their corresponding adjacent tokens before and after.

[0040] First, perform a similarity analysis on the token to be corrected and each token in the module corpus. The specific method is to calculate the similarity between each token in the corpus and the token to be corrected based on the edit distance, and then select the K most similar tokens in the corpus. In this embodiment, K = 5. Combine it with the adjacent tokens of the token to be corrected, that is, calculate the frequencies of the previous token and the similar tokens in the corpus that appear in the historical samples respectively, that is, the co-occurrence probability of the previous token and the similar tokens. At the same time, count the frequencies of the similar tokens in the corpus and the subsequent token that appear in the samples, that is, the co-occurrence probability of the subsequent token and the similar tokens; record the co-occurrence probability of the previous token and the similar tokens as the first frequency, and record the contribution probability of the subsequent token and the similar tokens as the second frequency. Determine the first possibility of the similar token based on the first frequency, the second frequency, and the edit distance between the token to be corrected and the similar token.

[0041] The first possibility is positively correlated with the first product and the second product respectively, and negatively correlated with the edit distance between the token to be corrected and the similar token.

[0042] It should be noted that negative correlation means that when one variable increases, the other variable decreases, and the change directions of the two variables are opposite. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small; the specific relationship is determined by the actual application, and this application does not make special restrictions.

[0043] Preferably, in this embodiment, the expression of the first possibility is: , represents the first product, represents the second product, represents the edit distance between the token to be corrected and the similar token, represents the first possibility of the similar token.

[0044] That is, if the edit distance between the similar segmentation and the segmentation to be corrected is smaller, and the frequency of co-occurrence of the adjacent front and rear segmentations of the similar segmentation and the segmentation to be corrected in the historical samples is higher, it indicates that the similar segmentation is more likely to be the content originally intended to be input by the segmentation to be corrected. If there is no other segmentation adjacent to the segmentation to be corrected (such as gender, marital status, etc.), then the reciprocal of the edit distance between the segmentation to be corrected and the similar segmentation is recorded as the first possibility that the similar segmentation is the segmentation to be corrected.

[0045] Thus, the first possibility that each similar segmentation is the segmentation to be corrected is obtained.

[0046] Step S004, determine the second possibility through the correction result corresponding to the similar segmentation history and the image similarity between the segmentation to be corrected and the similar segmentation.

[0047] Considering that when correcting abnormal segmentations in the input text, the writing characteristics of the corresponding segmentations should also be considered, that is, for the segmentations that are commonly misrecognized, the corresponding image texts during actual recognition should be similar; for example, if "错" is misrecognized as "醋", then all the images misrecognized as vinegar should also be close; therefore, the image corresponding to the segmentation to be corrected during recognition can be compared with the images recognized as the text to be corrected in the history, so as to increase the reliability of the correction result. Therefore, the text image of the segmentation to be corrected is read through ORC and compared with the segmentation images in the corpus, and the recognition result is further compared in combination with the writing habits of the corresponding historical segmentations, so as to increase the accuracy of segmentation correction.

[0048] Specifically, for each similar segmentation, obtain the segmentation before being corrected to this similar segmentation in the history, denoted as the corrected segmentation, and calculate the correction frequency of each corrected segmentation, where the correction frequency is the ratio of the number of each corrected segmentation to the number of all corrected segmentations. If the corrected segmentation includes the segmentation to be corrected, then the correction frequency of the corrected segmentation is the correction frequency of the segmentation to be corrected, otherwise the correction frequency of the segmentation to be corrected is 0.

[0049] Calculate the similarity between the text image of each corrected segmentation and the text image of the segmentation to be corrected, and take the maximum value of the similarity as the image similarity between the segmentation to be corrected and the similar segmentation. The calculation method of the similarity is to perform edge detection on the image, then perform shape context matching, and normalize the obtained similarity.

[0050] Obtain the second possibility that the similar segmentation is the segmentation to be corrected based on the correction frequency of the segmentation to be corrected and the image similarity between the segmentation to be corrected and the similar segmentation.

[0051] The second possibility is positively correlated with the image similarity and the correction frequency respectively.

[0052] Preferably, in this embodiment, the expression of the second possibility is: , Indicates the error correction frequency of the segmentation to be corrected. The image similarity between the segmented words to be corrected and the similar segments is used to represent the similarity between the segments. This indicates the second possibility that a similar word segment is the word segment to be corrected.

[0053] In other words, the higher the frequency of error correction in the historical segmentation process of the word to be corrected, and the higher the similarity between the text image of the word to be corrected and the text image of similar word segments that were mistakenly identified as other corresponding text images, then it indicates that the similar word segments are more likely to be the corrected word segments.

[0054] At this point, the second possibility of each similar word segment being a word to be corrected was obtained.

[0055] Step S005: Determine the total probability and the necessity of word segmentation correction by considering the first probability, the second probability, and the ambiguity, and complete the word segmentation correction.

[0056] The above steps yielded the first and second probabilities that each similar word segment could be the word to be corrected. The total probability of a similar word segment being the word to be corrected was then determined based on the first and second probabilities.

[0057] The total probability is positively correlated with the first probability and the second probability, respectively.

[0058] Preferably, in this embodiment, the total probability is the product of the first probability and the second probability.

[0059] Before correcting the segment to be corrected, this application first needs to calculate the necessity of correction for the segment to be corrected. When the necessity of correction is low, that is, the segment may be a newly added segment allowed by the module, no correction is performed, and only a warning is issued for the segment. At the same time, if the necessity of correction is high, the segment with the highest total probability with the segment to be corrected in the corpus of the corresponding module is selected as the segment after correction using the above method.

[0060] The necessity of word segmentation correction is positively correlated with the total probability and negatively correlated with the degree of ambiguity.

[0061] In this embodiment, the necessity of word segmentation correction is the normalized value of the ratio of total probability to ambiguity.

[0062] In other words, the higher the ambiguity of the text module containing the segment to be corrected, and the lower the similarity to the segment in the corpus, the lower the necessity for correction. The necessity for correction is compared with a correction threshold. If the necessity is less than the threshold, the segment may be new content allowed by the module or incorrect. In this case, the segment is marked and requires manual judgment. If the segment is deemed correct, it is added to the corpus. Conversely, if the necessity is high, the segment with the highest probability of similarity to the segment in the corresponding module's corpus is selected as the corrected segment. In this embodiment, the correction threshold is 0.3.

[0063] This completes the word segmentation error correction.

[0064] Based on the same inventive concept as the above methods, embodiments of the present invention also provide an intelligent form filling and error correction system for foreign affairs forms based on natural language processing, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described intelligent form filling and error correction methods for foreign affairs forms based on natural language processing.

[0065] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0066] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A foreign affairs table intelligent form filling and error correction method based on natural language processing, characterized by, The method comprises the following steps: Obtaining the text image of the corpus and vocabulary corresponding to each module of the foreign affairs form; After inputting the text, the input word is segmented to obtain the input word segmentation, and the corpus is used to screen the to-be-corrected word segmentation from the input word segmentation; for each module; the type of the input word segmentation is counted, and the ambiguity of each module is determined based on the frequency of the input word segmentation, the average frequency of all input word segmentations, and the type of the input word segmentation; Based on the edit distance of the to-be-corrected word segmentation and the word segmentation in the corpus, similar word segmentations are determined; the co-occurrence probability of the adjacent word segmentation and the similar word segmentation of the to-be-corrected word segmentation is calculated, and the first possibility of each similar word segmentation being the to-be-corrected word segmentation is determined based on the co-occurrence probability and the edit distance of the to-be-corrected word segmentation and the similar word segmentation; The correction frequency of the to-be-corrected word segmentation is determined based on the frequency of the word segmentation corrected to the similar word segmentation in the history; the second possibility of the similar word segmentation is obtained based on the correction frequency of the to-be-corrected word segmentation and the similarity of the text image of the to-be-corrected word segmentation and the similar word segmentation; The total possibility is determined based on the first possibility and the second possibility of the correction of the word segmentation; the necessity of the correction of the word segmentation is determined based on the total possibility and the ambiguity of the module; the word segmentation is corrected based on the necessity of the correction of the word segmentation and the total possibility.

2. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The frequency of the input word segmentation is the number of times of the appearance of the input word segmentation in the history divided by the number of times of the appearance of all input word segmentations.

3. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The ambiguity is positively correlated with the difference between the frequency of each input word segmentation and the average frequency of all input word segmentations, and the number of types of input word segmentations.

4. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The first possibility is positively correlated with the first product and the second product, and is negatively correlated with the edit distance of the to-be-corrected word segmentation and the similar word segmentation; the first product and the second product are the co-occurrence probability of the adjacent word segmentation and the similar word segmentation of the to-be-corrected word segmentation.

5. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The method for determining the correction frequency of the to-be-corrected word segmentation based on the frequency of the word segmentation corrected to the similar word segmentation in the history is as follows: For each similar word segmentation, the word segmentation before the correction to the similar word segmentation in the history is obtained, which is recorded as the corrected word segmentation, and the correction frequency of each corrected word segmentation is calculated, wherein the correction frequency is the ratio of the number of each corrected word segmentation to the number of all corrected word segmentations; if the corrected word segmentation includes the to-be-corrected word segmentation, the correction frequency of the corrected word segmentation is the correction frequency of the to-be-corrected word segmentation, otherwise the correction frequency of the to-be-corrected word segmentation is 0.

6. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The second possibility is positively correlated with the text image similarity and the correction frequency.

7. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The total possibility is positively correlated with the first possibility and the second possibility.

8. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The necessity of the correction of the word segmentation is positively correlated with the total possibility, and is negatively correlated with the ambiguity.

9. The natural language processing-based foreign affairs form intelligent filling and correction method according to claim 1, wherein, The method for correcting the word segmentation based on the necessity of the correction of the word segmentation and the total possibility is as follows: The necessity of the correction of the word segmentation is compared with the correction threshold; if the necessity of the correction of the word segmentation is less than the correction threshold, the to-be-corrected word segmentation is not corrected; if the necessity of the correction of the word segmentation is greater than the correction threshold, the correction is needed, and the similar word segmentation with the maximum total possibility is taken as the word segmentation after the correction of the to-be-corrected word segmentation.

10. An intelligent foreign affairs form filling and correcting system based on natural language processing, comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that, The processor executes the computer program to realize the steps of the foreign affairs form intelligent filling and correction method based on natural language processing according to any one of claims 1-9.

Images