Transcription error correction method, system and equipment for pinyin and Chinese character mixed text
By constructing a transcription error correction model, feature extraction and error correction processing of Pinyin Chinese characters mixed texts of younger students is solved, and the problem of insufficient error correction ability of Pinyin Chinese characters mixed texts in the existing technology is achieved, achieving more efficient transcription and error correction effects.
Patent Information
- Application Number
- CN202411934264.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-25
AI Technical Summary
The existing pinyin input method system has limited error correction ability when processing Pinyin and Chinese characters mixed texts of younger students, especially when dealing with pinyin abbreviations and typos, and it is difficult to flexibly adapt to the input methods of different students.
A pre-constructed transcription error correction model is adopted, including feature extraction module and text correction module, to perform feature extraction and error correction processing on the detected text to be tested, and transcription error correction results are generated. This model combines the multi-head self-attention mechanism and a fully connected feedforward network to achieve pinyin to Chinese characters and Chinese characters error correction through the embedding layer, feature extraction layer and text correction layer.
It improves the transcription accuracy and error correction efficiency of mixed texts of Pinyin and Chinese characters, can accurately identify and correct students' common Pinyin and Chinese characters writing errors, reduce teachers' correction burden, and improve students' writing accuracy.
Smart Images
Figure CN119886109A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of text processing, and in particular to a transcription error correction method, system and equipment for mixed pinyin and Chinese character text. Background Art
[0002] In Chinese learning and writing scenarios, especially in the writing process of young elementary school students, mixed input of pinyin and Chinese characters is a very common phenomenon. When students encounter unfamiliar or unwritten Chinese characters when writing, they usually replace them with pinyin. Therefore, text content mixed with pinyin and Chinese characters is often found in students' homework, essays, and other academic texts. This mixed writing mode of pinyin and Chinese characters increases the complexity of text processing because pinyin needs to be correctly transcribed into Chinese characters, and possible typos need to be corrected. This problem is particularly evident in young elementary school students because their Chinese character writing level is relatively low and they rely more on pinyin to assist expression.
[0003] Although current pinyin input method systems can provide error correction and Chinese character suggestions during the process of pinyin input, this is mainly for real-time input, and the ability to correct errors in existing pinyin-Chinese mixed text is limited. The error correction of these input method systems usually relies on fixed vocabulary and rules, and lacks flexibility in the recognition of typos and pinyin in pinyin-Chinese mixed input. In addition, these systems usually cannot fully handle the typos that are unique to students at different ages, because students' vocabulary and pinyin usage habits change greatly during the learning process.
[0004] In addition, the prior art also proposes a method for correcting mixed text of Pinyin and Chinese characters based on natural language processing (NLP). This type of method utilizes the semantic association between Pinyin and Chinese characters, and realizes the error correction of Pinyin to Chinese characters through lexical analysis, semantic analysis and the matching relationship between Pinyin and Chinese characters. These methods usually identify the possible Chinese characters corresponding to Pinyin by training models, and then use context semantics to judge the most appropriate Chinese characters. However, the existing methods have the following shortcomings: 1. It depends on pre-set rules and vocabulary, and it is difficult to flexibly adapt to the Pinyin input methods of different students, especially the Pinyin abbreviations and typos of young students; 2. The recognition ability of proper nouns in specific fields is insufficient, and it is difficult to meet the special vocabulary needs of students in homework and writing; 3. It is mainly for real-time input. The batch error correction ability for existing Pinyin and Chinese character mixed texts is limited; 4. The transcription accuracy of Pinyin and Chinese character mixed texts is not high, especially for the situation where Pinyin abbreviations or dialect Pinyin exist in the text. Therefore, in order to improve the transcription and error correction efficiency of Pinyin-Chinese mixed text in student writing scenarios, a new transcription and error correction method for Pinyin-Chinese mixed text is needed to more accurately and efficiently process the Pinyin-Chinese mixed text of young students. Summary of the invention
[0005] The technical problem to be solved by the present invention is to provide a transcription error correction method, system and equipment for mixed pinyin and Chinese characters text to solve the above technical problems.
[0006] The technical solution of the present invention to solve the above technical problems is as follows: a transcription error correction method for a mixed text of pinyin and Chinese characters, comprising: obtaining a text to be detected, wherein the text to be detected is a text that mixes pinyin and Chinese characters; performing transcription error correction processing on the text to be detected by using a pre-constructed transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents the result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected; wherein the transcription error correction model comprises a feature extraction module and a text correction module, wherein the feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module.
[0007] The beneficial effects of the present invention are as follows: the present method can accurately identify the pinyin part in the text to be detected through the transcription error correction model, and transcribe it into the corresponding Chinese characters, thereby improving the accuracy of the student's written text. For the common pinyin and Chinese character writing errors of students, an automatic error correction scheme can be provided, which can effectively identify typos and help students improve their ability to write pinyin and Chinese characters correctly. In summary, the present method can provide an effective transcription error correction scheme for the mixed use of pinyin and Chinese characters in student writing scenarios, improve students' writing accuracy, and reduce the burden on teachers in the text correction process.
[0008] Based on the above technical solution, the present invention can also be improved as follows.
[0009] Furthermore, the transcription error correction model also includes an embedding layer, and the output end of the embedding layer is connected to the input end of the feature extraction module.
[0010] Furthermore, the text to be detected is transcribed and corrected by using a pre-constructed transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, including: digitizing the text to be detected by using the embedding layer to generate a coding vector corresponding to the text to be detected; performing feature extraction by using the feature extraction module based on the coding vector corresponding to the text to be detected to generate a feature vector corresponding to the text to be detected; and generating a transcription error correction result corresponding to the text to be detected by using the text correction module based on the feature vector corresponding to the text to be detected.
[0011] Furthermore, the feature extraction module includes six feature extraction layers, each of which includes a first fully connected feedforward network, a one-dimensional convolution sublayer and an attention sublayer connected in sequence.
[0012] Furthermore, the text correction module includes six text correction layers, each of which includes a connected multi-head self-attention sublayer and a second fully connected feed-forward network.
[0013] Furthermore, the calculation formula of the total loss function of the transcription error correction model is:
[0014] Loss_total=a*loss_pinyin+b*loss_char+(1-ab)*loss_correct;
[0015] Among them, loss_total is the total loss function, loss_pinyin is the pinyin loss function, the pinyin loss function is a loss function that reflects the degree of deviation of the feature extraction module in pinyin recognition, loss_char is the typo loss function, the typo loss function is a loss function that reflects the degree of deviation of the feature extraction module in typo recognition, loss_correct is the correctness loss function, the correctness loss function is a loss function that reflects the degree of deviation of the text correction module in correcting the recognized pinyin and typos, a is the first balancing factor, and b is the second balancing factor.
[0016] Furthermore, the pinyin loss function, the typo loss function and the correctness loss function are all cross entropy loss functions.
[0017] In order to solve the above technical problems, the present invention also proposes a transcription error correction system for a pinyin-Chinese character mixed text, comprising:
[0018] A text acquisition module, used to acquire a text to be detected, wherein the text to be detected is a text that uses a mixture of pinyin and Chinese characters;
[0019] A transcription error correction module, used for performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents a result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected;
[0020] A model building module is used to build the transcription error correction model, and the transcription error correction model includes a feature extraction module and a text correction module. The feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module.
[0021] In order to solve the above technical problems, the present invention also proposes an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the transcription error correction method for mixed pinyin and Chinese characters text as described above is implemented.
[0022] In order to solve the above technical problems, the present invention also proposes a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute a transcription and error correction method for a mixed text of pinyin and Chinese characters as described above. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 A flowchart of a transcription error correction method for a pinyin-Chinese character mixed text according to the present invention;
[0024] Figure 2 A schematic diagram of the structure of a transcription error correction model of a transcription error correction method for a pinyin-Chinese character mixed text according to the present invention;
[0025] Figure 3 A first fully connected feedforward network calculation graph of a transcription error correction method for a pinyin-Chinese character mixed text according to the present invention;
[0026] Figure 4 The present invention is a schematic diagram of a transcription error correction system for a pinyin-Chinese character mixed text. DETAILED DESCRIPTION
[0027] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.
[0028] like Figure 1 As shown, this embodiment provides a transcription error correction method for a pinyin-Chinese character mixed text, comprising:
[0029] S101, obtaining a text to be detected, wherein the text to be detected is a text that uses a mixture of pinyin and Chinese characters.
[0030] S102, performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents a result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected.
[0031] The transcription error correction model includes a feature extraction module and a text correction module. The feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module. Figure 2 shown.
[0032] Optionally, in an embodiment, the transcription error correction model further includes an embedding layer, and an output end of the embedding layer is connected to an input end of the feature extraction module.
[0033] Optionally, in an embodiment, the text to be detected is subjected to transcription error correction processing by using a pre-constructed transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, including: digitizing the text to be detected by using the embedding layer to generate a coding vector corresponding to the text to be detected; performing feature extraction by using the feature extraction module according to the coding vector corresponding to the text to be detected to generate a feature vector corresponding to the text to be detected; and generating a transcription error correction result corresponding to the text to be detected by using the text correction module according to the feature vector corresponding to the text to be detected.
[0034] Digitization is the process of converting text data into a numerical form that can be understood by machines, so that computers can perform in-depth processing on the text through numerical calculations and analysis. After the text to be detected is digitized, a coding vector is generated. The text to be detected is represented in the form of a vector. The coding vector can, to a certain extent, express the semantic information between two characters or words through methods such as cosine distance.
[0035] Optionally, in an embodiment, the feature extraction module includes six feature extraction layers, each of which includes a first fully connected feedforward network, a one-dimensional convolution sublayer and an attention sublayer connected in sequence.
[0036] Based on the text features of mixed pinyin and Chinese characters, this embodiment proposes a feature extraction module consisting of a fully connected, local feature extraction convolution layer, and a global feature extraction attentiveness, namely Pinyin-Char-Encoder. The feature extraction module is stacked by six identical feature extraction layers, each of which includes three sublayers: a first fully connected feedforward network, a one-dimensional convolution sublayer, and an attention sublayer, and each sublayer is added with residual connection (res idual connection ion) and normalization (normalization ion).
[0037] The first fully connected feedforward network is responsible for integrating the features extracted by the previous layers, mainly for data representation of single Chinese characters or pinyin. Figure 3As shown in the figure, the encoding vector X obtained after the embedding layer processing is input into the first fully connected feed-forward network (FFN), and the first fully connected feed-forward network performs vector-level calculations on X. Specifically, this is to perform operations such as matrix multiplication and dot product on the input vector X. The result is a vector representation with the same dimension as the encoding vector, but this vector is closer to the final task.
[0038] The one-dimensional convolutional layer is used to extract local features from the input data (the output of the first fully connected feedforward network), that is, to perform better feature extraction and data representation for adjacent Chinese characters and pinyin data. The definition of the one-dimensional convolutional layer (Conv-1D) is:
[0039]
[0040] Among them, u represents the position index in the output sequence, corresponding to the position after the convolution operation. v represents the channel index, which is used in multi-channel convolution and corresponds to the output channel. z(u,v) represents the value of the convolution output.
[0041] i represents the position index in the input sequence, which is used to traverse the input data. j represents the channel index of the input data, which is used to traverse the channels of the input data. i,j Represents input data.
[0042] k u-i,v-j Represents the weight of the convolution kernel (or filter), where ui is the position offset of the convolution kernel on the input sequence, and vj is the offset of the convolution kernel in the channel dimension.
[0043] In practical applications, ∞ and -∞ represent the sliding range of the convolution kernel on the input data. In actual operation, the convolution kernel usually only slides within the valid range of the input data, rather than on the entire infinite sequence.
[0044] The attention sublayer enables the model to capture global information, focus on key information, and ignore irrelevant information. The definition of the attention sublayer is:
[0045] MultiHead(Q,K,V)=Concat(head1,head2,…,headn)Wo;
[0046] headi=Self_attention(Q*W1,K*W2,V*W3);
[0047]
[0048] Among them, MultiHead represents the multi-head mechanism, Concat represents the concatenation operation, headi represents the calculation method of the i-th head, Self_attention represents the self-attention mechanism, Wo is the output mapping matrix, and d k is the dimension of K vector. Multi-head attention projects Q, K, V through n different linear transformations, and finally concatenates the unused attention results. Q, K, V are vectors obtained by multiplying the input vector with different matrices W1, W2, and W3 respectively, and then the Q, K, and V matrices composed of several vectors are convenient for simultaneous operations.
[0049] Optionally, in an embodiment, the text correction module includes six text correction layers, each of which includes a connected multi-head self-attention sublayer and a second fully connected feed-forward network.
[0050] The text correction module (Transformer decoder) is composed of six identical text correction layers stacked together. Each layer includes two sublayers: a multi-head self-attention sublayer and a second fully connected feedforward network, and each sublayer has residual connections and normalization. The multi-head self-attention sublayer enables the model to capture global information, focus on key information, and ignore irrelevant information. The second fully connected feedforward network is responsible for integrating the features extracted by the previous layer, mainly for data representation of single Chinese characters or pinyin.
[0051] The output of the Transformer encoder can be expressed as:
[0052] Transforme_encoder=LayerNorm(X+(SubLayer(X)));
[0053] Among them, LayerNorm represents the normalization of the layer, SubLayer represents the sublayer, and X represents the encoding vector.
[0054] For the training of the transcription error correction model: First, data collection is carried out. A large amount of mixed texts of Pinyin and Chinese characters are collected from students' writing assignments and essays, including Pinyin abbreviations, mixed Pinyin and Chinese characters, typos, and other types. The collected texts are cleaned to remove unnecessary special characters and standardize the mixed format of Pinyin and Chinese characters for model training. Among them, unnecessary special characters refer to smears that appear in the actual writing process, small mark word count prompts in the composition grid, etc. The initial transcription error correction model is trained with the cleaned text, and the structure of the initial transcription error correction model is the same as described above.
[0055] First, we pre-trained the model on a large-scale bilingual corpus of Chinese characters and pinyin to master the basic mapping relationship between pinyin and Chinese characters. Then we fine-tuned the model for specific texts in student writing scenarios to enhance the model's ability to recognize and correct pinyin abbreviations, pinyin-Chinese character mixtures, and typos. In addition, we used optimization strategies such as learning rate adjustment and gradient clipping to improve the model's training efficiency and generalization ability.
[0056] This embodiment proposes a multi-task joint loss function based on multiple tasks of pinyin recognition, typo recognition and transcription correction.
[0057] Optionally, in an embodiment, the calculation formula of the total loss function of the transcription error correction model is:
[0058] Loss_total=a*loss_pinyin+b*loss_char+(1-ab)*loss_correct;
[0059] Among them, loss_total is the total loss function, loss_pinyin is the pinyin loss function, the pinyin loss function is a loss function that reflects the degree of deviation of the feature extraction module in pinyin recognition, loss_char is the typo loss function, the typo loss function is a loss function that reflects the degree of deviation of the feature extraction module in typo recognition, loss_correct is the correctness loss function, the correctness loss function is a loss function that reflects the degree of deviation of the text correction module in correcting the recognized pinyin and typos, a is the first balancing factor, and b is the second balancing factor.
[0060] By setting the first balancing factor and the second balancing factor, the weights of the pinyin loss function, the typo loss function, and the correctness loss function are determined. loss_pinyin is used to determine whether each character in the text to be detected is pinyin, loss_char is used to determine whether each character in the text to be detected is a typo, and loss_correct is used for transcription and error correction tasks.
[0061] Optionally, in an embodiment, the pinyin loss function, the typo loss function and the correctness loss function are all cross entropy loss functions.
[0062] This method proposes an efficient transcription and error correction method based on the characteristics of mixed texts of pinyin and Chinese characters in student writing scenarios, aiming to solve the current difficulties in processing mixed texts of pinyin and Chinese characters. For the pinyin in the text, the transcription error correction model generates a possible set of Chinese character candidates and selects the most suitable Chinese characters based on context information. For texts that already contain Chinese characters, the model identifies and corrects typos based on context and vocabulary information.
[0063] This method is applicable to all educational scenarios involving the mixed use of pinyin and Chinese characters. The trained transcription and error correction model is embedded in a pinyin-Chinese mixed text processing system, providing pinyin-to-Chinese character conversion, typo correction and batch processing functions to meet the needs of students' writing scenarios.
[0064] In summary, this method can accurately identify the pinyin part of the text through advanced natural language processing technology, and transcribe it into corresponding Chinese characters, thereby improving the accuracy of students' written texts. For students' common pinyin and Chinese character input errors, it can provide automatic error correction solutions, effectively identify typos, and help students improve their ability to write pinyin and Chinese characters correctly. During use, the transcription error correction model can be appropriately optimized and adjusted, and the transcription and error correction strategies can be flexibly adjusted according to the input habits and usage scenarios of students of different age groups to provide more targeted text processing solutions. It also supports dynamic vocabulary changes in students' learning process, updates the vocabulary library in a timely manner, and identifies subject-specific nouns and emerging vocabulary to improve the accuracy of transcription and error correction. It also provides batch processing functions, which can quickly transcribe and correct a large number of mixed texts of pinyin and Chinese characters, and improve the efficiency of teachers in marking homework and processing texts.
[0065] Embodiment 3
[0066] like Figure 4 As shown, this embodiment provides a transcription error correction system for a pinyin-Chinese mixed text, including:
[0067] A text acquisition module, used to acquire a text to be detected, wherein the text to be detected is a text that uses a mixture of pinyin and Chinese characters;
[0068] A transcription error correction module, used for performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents a result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected;
[0069] A model building module is used to build the transcription error correction model, and the transcription error correction model includes a feature extraction module and a text correction module. The feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module.
[0070] Optionally, in an embodiment, the transcription error correction model further includes an embedding layer, and an output end of the embedding layer is connected to an input end of the feature extraction module.
[0071] Optionally, in an embodiment, the transcription error correction module includes:
[0072] A digitizing unit, used for digitizing the text to be detected through the embedding layer to generate a coding vector corresponding to the text to be detected;
[0073] A feature extraction unit, used to perform feature extraction through the feature extraction module according to the encoding vector corresponding to the text to be detected, and generate a feature vector corresponding to the text to be detected;
[0074] The transcription error correction unit is used to generate a transcription error correction result corresponding to the text to be detected through the text correction module according to the feature vector corresponding to the text to be detected.
[0075] Optionally, in an embodiment, the feature extraction module includes six feature extraction layers, each of which includes a first fully connected feedforward network, a one-dimensional convolution sublayer and an attention sublayer connected in sequence.
[0076] Optionally, in an embodiment, the text correction module includes six text correction layers, each of which includes a connected multi-head self-attention sublayer and a second fully connected feed-forward network.
[0077] Optionally, in an embodiment, the calculation formula of the total loss function of the transcription error correction model is:
[0078] Loss_total=a*loss_pinyin+b*loss_char+(1-ab)*loss_correct;
[0079] Among them, loss_total is the total loss function, loss_pinyin is the pinyin loss function, the pinyin loss function is a loss function that reflects the degree of deviation of the feature extraction module in pinyin recognition, loss_char is the typo loss function, the typo loss function is a loss function that reflects the degree of deviation of the feature extraction module in typo recognition, loss_correct is the correctness loss function, the correctness loss function is a loss function that reflects the degree of deviation of the text correction module in correcting the recognized pinyin and typos, a is the first balancing factor, and b is the second balancing factor.
[0080] Optionally, in an embodiment, the pinyin loss function, the typo loss function and the correctness loss function are all cross entropy loss functions.
[0081] Embodiment 3
[0082] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a transcription error correction method for a mixed text of pinyin and Chinese characters as described in Embodiment 1 is implemented.
[0083] Embodiment 4
[0084] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute a transcription error correction method for a pinyin-Chinese character mixed text as described in Example 1.
[0085] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0086] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.
Claims
1. A transcription error correction method for a mixed text of pinyin and Chinese characters, characterized in that: include: Obtaining a text to be detected, wherein the text to be detected is a text that uses a mixture of pinyin and Chinese characters; Performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents a result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected; Among them, the transcription error correction model includes a feature extraction module and a text correction module, the feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module.
2. According to the transcription error correction method of a pinyin-Chinese character mixed text as described in claim 1, it is characterized in that: The transcription error correction model also includes an embedding layer, and an output end of the embedding layer is connected to an input end of the feature extraction module.
3. According to the transcription error correction method of a pinyin-Chinese character mixed text as described in claim 2, it is characterized in that: The method of performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected includes: Digitally process the text to be detected through the embedding layer to generate a coding vector corresponding to the text to be detected; According to the encoding vector corresponding to the text to be detected, the feature extraction module performs feature extraction to generate a feature vector corresponding to the text to be detected; According to the feature vector corresponding to the text to be detected, the transcription error correction result corresponding to the text to be detected is generated by the text correction module.
4. The transcription error correction method for a mixed text of phonetic and Chinese characters according to claim 1, characterized in that: The feature extraction module includes six feature extraction layers, each of which includes a first fully connected feedforward network, a one-dimensional convolution sublayer and an attention sublayer connected in sequence.
5. The transcription error correction method for a mixed text of phonetic and Chinese characters according to claim 1, characterized in that: The text correction module includes six text correction layers, each of which includes a connected multi-head self-attention sublayer and a second fully connected feed-forward network.
6. The transcription error correction method for a mixed text of phonetic and Chinese characters according to claim 1, characterized in that: The calculation formula of the total loss function of the transcription error correction model is: Loss_total=a*loss_pinyin+b*loss_char+(1-ab)*loss_correct; Among them, loss_total is the total loss function, loss_pinyin is the pinyin loss function, the pinyin loss function is a loss function that reflects the degree of deviation of the feature extraction module in pinyin recognition, loss_char is the typo loss function, the typo loss function is a loss function that reflects the degree of deviation of the feature extraction module in typo recognition, loss_correct is the correctness loss function, the correctness loss function is a loss function that reflects the degree of deviation of the text correction module in correcting the recognized pinyin and typos, a is the first balancing factor, and b is the second balancing factor.
7. The transcription error correction method for a mixed text of phonetic and Chinese characters according to claim 6, characterized in that: The pinyin loss function, the typo loss function and the correctness loss function are all cross entropy loss functions.
8. A transcription error correction system for mixed pinyin and Chinese characters text, characterized in that: include: A text acquisition module, used to acquire a text to be detected, wherein the text to be detected is a text that uses a mixture of pinyin and Chinese characters; A transcription error correction module, used for performing transcription error correction processing on the text to be detected by using a pre-built transcription error correction model to obtain a transcription error correction result corresponding to the text to be detected, wherein the transcription error correction result represents a result after the pinyin in the text to be detected is converted into Chinese characters and the Chinese characters in the text to be detected are corrected; A model building module is used to build the transcription error correction model, and the transcription error correction model includes a feature extraction module and a text correction module. The feature extraction module is used to extract features of the text to be detected, and the text correction module is used to generate the transcription error correction result based on the features extracted by the feature extraction module.
9. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a transcription error correction method for a pinyin-Chinese character mixed text as described in any one of claims 1 to 7 is implemented.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute a transcription error correction method for a mixed pinyin and Chinese character text as described in any one of claims 1 to 7.