Japanese writing automatic error correction method and system

Abstract

The present invention discloses a method and system for automatically correcting errors in Japanese writing. The method comprises the following steps: converting each word in a Japanese writing text to be detected into a word vector, extracting features from the word vectors in each sentence in the text, obtaining a word vector with features in the sentence, and using the word vectors as feature quantities, inputting the word vectors and feature quantities in the sentence into an association model, performing learning and adjustment on the association model using a feed-back propagation method, obtaining an association value between each word vector and the feature quantity in the sentence, obtaining a set of combinations between word vectors and word vectors according to parts of speech, selecting a group of combinations with the largest total association value in the set, marking unselected word vectors in the combination, and improving the selected combination according to a knowledge graph to generate a text after error correction. The present invention can play the role of automatic error correction, help writers automatically detect text errors, save review time, and also improve writing efficiency.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for automatically correcting errors in Japanese writing. Background Art

[0002] In recent years, with the deepening of my country's opening up to the outside world, the demand for talents proficient in minority languages has been growing. Japan, as a neighboring economy, has become increasingly closely connected with my country. Economic activities require complex communication, making Japanese a popular language. This trend has driven people's enthusiasm for learning Japanese, and the number of students studying Japanese has been rising sharply each year. However, this rapid development has also increased the pressure on teachers. Due to the scarcity of teachers, students need to have a certain ability to self-check. However, students currently lack this ability to self-check and correct errors in the writing process, and the error correction process is not scientific enough, wasting time and reducing learning efficiency. Summary of the Invention

[0003] In order to at least solve or partially solve the above problems, a method and system for automatic error correction in Japanese writing are provided.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] The present invention provides a Japanese writing automatic error correction method, comprising the following steps:

[0006] S1: Collect Japanese writing texts to be tested;

[0007] S2: Convert each word in the Japanese writing text to be tested into a word vector. Perform feature extraction on the word vectors in each sentence in the text to obtain a word vector with characteristics in the sentence, and use it as a feature quantity. Input the word vectors and feature quantities in the sentence into the association model. The association model uses feedback propagation to perform learning and adjustment to obtain the association value between each word vector and the feature quantity in the sentence. According to the part of speech, a set of combinations between word vectors and word vectors is obtained. The set of combinations with the largest total association value in the set is selected, and the unselected word vectors in the combination are marked.

[0008] S3: Improve the selected combination according to the knowledge tree to generate the corrected text.

[0009] As a preferred technical solution of the present invention, in step S2, the association model is modeled according to the database through a deep learning algorithm to obtain the association probability between the feature quantity and the word vector and the association probability between the word vector and the feature quantity, and the two association probabilities are divided to obtain the association value.

[0010] As an optimal technical solution of the present invention, the deep learning is carried out using a convolutional neural network, which is composed of an encoder path and a decoder path. Both the encoder path and the decoder path are composed of a network structure in the form of a convolution unit from convolution to batch normalization to activation of a ReLU function.

[0011] As a preferred technical solution of the present invention, in step S3, the word vectors selected in the combination are used as nodes, and a Monte Carlo tree is used to search for the missing word vectors in the combination to form an error-correcting text.

[0012] The present invention provides an automatic error correction system for Japanese writing, comprising a database, a word vector conversion module, a feature extraction module, a text acquisition module, an association model, a text selection module, an error marking module, and an error correction module;

[0013] The text acquisition module is used to acquire the Japanese writing text to be detected;

[0014] The word vector conversion module converts the acquired Japanese writing text into a word vector;

[0015] The feature extraction module extracts the most characteristic word vector of each sentence in the text as a feature quantity;

[0016] The association model is modeled according to the database, and the word vectors and feature quantities in the sentence are adjusted by a deep learning algorithm and a feed-back propagation method to obtain the association value between each word vector and feature quantity in the sentence;

[0017] The text selection module obtains a set of word vectors and feature quantity combinations according to the word part of speech, uses a greedy algorithm to obtain a combination with the largest correlation value in the set, and selects the combination;

[0018] The error marking module marks the word vectors not selected in the combination to generate text with error marks;

[0019] The error correction module improves the selected combination through the knowledge tree to generate a corrected text.

[0020] As a preferred technical solution of the present invention, it also includes a display module, which compares and displays the unselected word vectors and the added and improved word vectors.

[0021] As a preferred technical solution of the present invention, it also includes a statistical module, which counts the error rate in writing and generates a score.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention can play a role in automatic error correction, helping writers to automatically detect text errors, saving time for review, and also improving writing efficiency, ensuring writing quality while improving the author's writing level. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 is a flow chart of the method of the present invention;

[0026] Figure 2 It is the overall system structure diagram of the present invention;

[0027] In the figure: 1. Text acquisition module; 2. Word vector conversion module; 3. Feature extraction module; 4. Association model; 5. Database; 6. Text selection module; 7. Error marking module; 8. Error correction module; 9. Display module; 10. Statistics module. DETAILED DESCRIPTION

[0028] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention. In the accompanying drawings, the same reference numerals all refer to the same components.

[0029] Furthermore, if detailed description of known art is not necessary to illustrate the characteristics of the present invention, it will be omitted.

[0030] Example 1

[0031] like Figure 1 As shown, the present invention provides a method for automatic error correction of Japanese writing, comprising the following steps:

[0032] S1: Collect Japanese writing texts to be tested;

[0033] S2: Convert each word in the Japanese writing text to be tested into a word vector. Perform feature extraction on the word vectors in each sentence in the text to obtain a word vector with characteristics in the sentence, and use it as a feature quantity. Input the word vectors and feature quantities in the sentence into the association model. The association model uses feedback propagation to perform learning and adjustment to obtain the association value between each word vector and the feature quantity in the sentence. According to the part of speech, a set of combinations between word vectors and word vectors is obtained. The set of combinations with the largest total association value in the set is selected, and the unselected word vectors in the combination are marked.

[0034] S3: Improve the selected combination according to the knowledge tree to generate the corrected text.

[0035] Specifically, the main information in the sentence, such as the main meaning of the expression in the sentence, is used, and the correlation between the main information and the words is used to detect whether the writing is wrong. In the process of calculating the correlation between the main information and the words, we first use feature extraction to extract the key information in the sentence, and divide the correlation probability between the correlation information and the words and the correlation probability between the words and the correlation information to obtain the correlation value. The correlation probability between the feature quantity and the word vector and the correlation probability between the word vector and the feature quantity are completed using a convolutional neural network. The convolutional neural network consists of an encoder path and a decoder path. The encoder path and the decoder path are both composed of a network structure in the form of a convolution unit from convolution to batch normalization to activation of the Relu function. The encoder path uses a convolutional layer to generate encoder data, and then uses batch normalization for normalization and uses the ReLU activation function to activate it. The decoder path uses a deconvolution layer to decode the data generated by the encoder, and uses batch normalization for normalization and uses the ReLU activation function to activate it.

[0036] After obtaining the association value between each word and the feature value, we combine them according to the part of speech, which refers to verbs, nouns, etc., and then find the combination with the largest association value. This combination is the content expressed by the author. Due to the relationship between the parts of speech, if there are errors in the writing, some words will conflict with each other, and unselected word vectors will appear. If there are no errors, all words will be selected, and then the unselected words will be marked, and the word vectors in the combination will be improved. During the improvement process, the selected word vectors in the combination are used as nodes, and the Monte Carlo tree is used to search to improve the missing word vectors in the combination to form an error-corrected text.

[0037] According to the above method, if Figure 2 As shown, the present invention also provides a Japanese writing automatic error correction system, comprising a database 5, a word vector conversion module 2, a feature extraction module 3, a text acquisition module 1, an association model 4, a text selection module 6, an error marking module 7, and an error correction module 8;

[0038] The text acquisition module 1 is used to acquire the Japanese writing text to be detected;

[0039] The word vector conversion module 2 converts the acquired Japanese writing text into a word vector;

[0040] Feature extraction module 3 extracts the most characteristic word vector of each sentence in the text as the feature quantity;

[0041] The association model 4 is modeled based on the database 5. The word vectors and feature quantities in the sentence are adjusted using a deep learning algorithm and a feed-back propagation method to obtain the association value between each word vector and feature quantity in the sentence.

[0042] The text selection module 6 obtains a set of word vectors and feature quantity combinations according to the word part of speech, uses a greedy algorithm to obtain the combination with the largest correlation value in the set, and selects the combination;

[0043] The error marking module 7 marks the word vectors that are not selected in the combination and generates text with error marks;

[0044] The error correction module 8 improves the selected combination through the knowledge tree to generate a corrected text.

[0045] In order to facilitate the review of errors by the writers, the display module 9 is also used to compare and display the unselected word vectors and the added and improved word vectors. The comparison effect can make the writers aware of their own errors.

[0046] At the same time, the statistical module 10 is used to collect statistics on the error rate in writing and generate a score, so that the student can accurately know his or her writing level.

[0047] The present invention can play a role in automatic error correction, helping writers to automatically detect text errors, saving time for review, and also improving writing efficiency, ensuring writing quality while improving the author's writing level.

[0048] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for automatic error correction in Japanese writing, characterized in that: The following steps are involved: S1: Collect Japanese writing texts to be tested; S2: converting each word in the Japanese writing text to be detected into a word vector, extracting features from the word vectors in each sentence in the text, obtaining a word vector with features in the sentence, and using it as a feature quantity, inputting the word vector and the feature quantity in the sentence into an association model, and the association model is learned and adjusted by feedback propagation to obtain an association value between each word vector in the sentence and the feature quantity, obtaining a set of combinations between word vectors and word vectors according to parts of speech, selecting a set of combinations with the largest total association value in the set, and marking the word vectors that are not selected in the combination; in the step S2, the association model is modeled according to the database through a deep learning algorithm to obtain an association probability between the feature quantity and the word vector and an association probability between the word vector and the feature quantity, and dividing the two association probabilities to obtain an association value; S3: The selected combination is improved according to the knowledge tree to generate a corrected text; in step S3, the word vectors selected in the combination are used as nodes, and a Monte Carlo tree is searched to improve the missing word vectors in the combination to form a corrected text.

2. The Japanese writing automatic error correction method according to claim 1, characterized in that: The deep learning is accomplished using a convolutional neural network, which consists of an encoder path and a decoder path, both of which consist of a network structure in the form of convolutional units from convolution to batch normalization to activation of a Relu function.

3. A Japanese writing automatic error correction system, characterized in that: It includes a database (5), a word vector conversion module (2), a feature extraction module (3), a text acquisition module (1), an association model (4), a text selection module (6), an error marking module (7), and an error correction module (8); The text acquisition module (1) is used to acquire the Japanese writing text to be detected; The word vector conversion module (2) converts the acquired Japanese writing text into a word vector; The feature extraction module (3) extracts the most characteristic word vector from each sentence in the text as a feature; The association model (4) is modeled according to the database (5), and the word vectors and feature quantities in the sentence are adjusted by a deep learning algorithm and a feed-back propagation method to obtain the association probability between the feature quantity and the word vector and the association probability between the word vector and the feature quantity, and the association value is obtained by dividing the two association probabilities; The text selection module (6) obtains a set of word vectors and feature quantity combinations according to the word part of speech, uses a greedy algorithm to obtain a combination with the largest correlation value in the set, and selects the combination; The error marking module (7) marks the word vectors not selected in the combination to generate text with error marks; The error correction module (8) uses the word vectors selected in the combination as nodes, searches through the Monte Carlo tree, and improves the word vectors missing in the combination to form an error correction text.

4. The Japanese writing automatic error correction system according to claim 3, characterized in that: The method further comprises a display module (9), wherein the display module (9) compares and displays the unselected word vectors and the added and improved word vectors.

5. The Japanese writing automatic error correction system according to claim 3, characterized in that: The system further includes a statistical module (10), which counts the error rate in writing and generates a score.

Images