A text error correction method, device, computer equipment and storage medium

Abstract

The present invention provides a text error correction method, apparatus, computer equipment and storage medium, which relate to the field of computer technology. The method can determine a secondary truth label corresponding to a primary truth label of a training sample, and train a text error correction model based on the primary truth label and the secondary truth label, so that the probability of the secondary truth label of the training sample determined by the trained text error correction model as a true value result is less than the probability of the primary truth label as a true value result, and greater than the probability of a non-true value label as a true value result, thereby avoiding the overfitting disadvantage caused by a single correct label during training and reducing false detection.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a text error correction method, apparatus, computer equipment and storage medium. Background Art

[0002] Text error correction refers to the process of automatically detecting and correcting errors in text. These errors may include spelling errors, grammatical errors, word errors, etc. Text error correction technology plays an important role in the field of natural language processing and is widely used in document editing, search engines, speech recognition and other fields to improve text quality and user experience.

[0003] Text error correction is usually based on a pre-trained model. In the model's training data, the correct labels of the samples are unique, which can easily lead to model overfitting and false detection. Summary of the invention

[0004] The embodiments of the present disclosure at least provide a text error correction method, apparatus, computer equipment and storage medium.

[0005] In a first aspect, an embodiment of the present disclosure provides a text error correction method, comprising:

[0006] Constructing a text error correction model to be trained, and obtaining a plurality of training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character;

[0007] Determining at least one secondary truth label that matches the primary truth label of the training sample;

[0008] The text error correction model to be trained is trained using the training sample carrying the preferred true value label and the secondary true value label corresponding to the training sample to obtain a trained text error correction model; the trained text error correction model makes the probability that the secondary true value label of the training sample is the true value result less than the probability that the preferred true value label is the true value result, and greater than the probability that the non-true value label is the true value result;

[0009] The text to be corrected is obtained, and the text to be corrected is input into the trained text correction model to obtain the correction result of the text to be corrected.

[0010] In an optional implementation, the determining at least one secondary truth label that matches the primary truth label of the training sample includes:

[0011] For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence;

[0012] Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence;

[0013] Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies;

[0014] Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

[0015] In an optional implementation manner, the generating, based on the context information of the erroneous character in the training sample sentence, a first character sequence containing the context information and the to-be-determined character includes:

[0016] Determine a first adjacent character before the position where the erroneous character is located in the training sample sentence, and a second adjacent character after the position where the erroneous character is located in the training sample sentence;

[0017] A first character sequence is generated based on the first adjacent character and the second adjacent character; the first character sequence includes the first adjacent character, the to-be-determined character, and the second adjacent character arranged in sequence.

[0018] In an optional implementation manner, determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies includes:

[0019] Determine, from the second character sequence, a fourth character sequence whose occurrence frequency is greater than or equal to a preset frequency;

[0020] sorting the fourth character sequence according to the occurrence frequency;

[0021] The fourth character sequences of the first preset number with the highest occurrence frequency are determined as the third character sequence.

[0022] In an optional implementation, the training sample carrying the first true value label and the second true value label corresponding to the training sample are used to train the text error correction model to obtain the trained text error correction model, including:

[0023] Inputting the training sample into the text error correction model to be trained, obtaining prediction results of multiple target characters corresponding to the erroneous character determined by the text error correction model to be trained; the prediction results represent the probability that the correct character corresponding to the erroneous character is the target character;

[0024] Determine a first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label;

[0025] Determine a second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label;

[0026] determining a third loss according to the sum of the first loss and the second loss;

[0027] The text error correction model to be trained is trained using the third loss to obtain a trained text error correction model.

[0028] In an optional implementation manner, determining the first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label includes:

[0029] For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter;

[0030] determining a second correction value based on a product of the first correction value and a second correction parameter;

[0031] The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

[0032] In an optional implementation, determining the second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label includes:

[0033] For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter;

[0034] determining a fourth correction value based on a product between the third correction value and a fourth correction parameter;

[0035] The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

[0036] In an optional implementation manner, the using the third loss to train the text error correction model to be trained to obtain a trained text error correction model includes:

[0037] The parameters of the text error correction model to be trained are adjusted so that the third loss is lower than or equal to the target loss threshold, thereby obtaining a trained text error correction model.

[0038] In a second aspect, the present disclosure also provides a text error correction device, including:

[0039] A construction module is used to construct a text error correction model to be trained, and obtain multiple training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character;

[0040] A determination module, configured to determine at least one secondary truth label that matches the primary truth label of the training sample;

[0041] A training module, used to train the text error correction model to be trained using the training samples carrying the preferred true value labels and the secondary true value labels corresponding to the training samples, to obtain a trained text error correction model; the trained text error correction model makes the probability of the secondary true value label of the training sample being the true value result less than the probability of the preferred true value label being the true value result, and greater than the probability of the non-true value label being the true value result;

[0042] The error correction module is used to obtain the text to be corrected, input the text to be corrected into the trained text error correction model, and obtain the error correction result of the text to be corrected.

[0043] In an optional implementation manner, the determining module is specifically configured to:

[0044] For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence;

[0045] Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence;

[0046] Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies;

[0047] Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

[0048] In an optional implementation manner, the determining module is specifically configured to:

[0049] Determine a first adjacent character before the position where the erroneous character is located in the training sample sentence, and a second adjacent character after the position where the erroneous character is located in the training sample sentence;

[0050] A first character sequence is generated based on the first adjacent character and the second adjacent character; the first character sequence includes the first adjacent character, the to-be-determined character, and the second adjacent character arranged in sequence.

[0051] In an optional implementation manner, the determining module is specifically configured to:

[0052] Determine, from the second character sequence, a fourth character sequence whose occurrence frequency is greater than or equal to a preset frequency;

[0053] sorting the fourth character sequence according to the occurrence frequency;

[0054] The fourth character sequences of the first preset number with the highest occurrence frequency are determined as the third character sequence.

[0055] In an optional implementation manner, the training module is specifically used for:

[0056] Inputting the training sample into the text error correction model to be trained, obtaining prediction results of multiple target characters corresponding to the erroneous character determined by the text error correction model to be trained; the prediction results represent the probability that the correct character corresponding to the erroneous character is the target character;

[0057] Determine a first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label;

[0058] Determine a second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label;

[0059] determining a third loss according to the sum of the first loss and the second loss;

[0060] The text error correction model to be trained is trained using the third loss to obtain a trained text error correction model.

[0061] In an optional implementation manner, the training module is specifically used for:

[0062] For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter;

[0063] determining a second correction value based on a product of the first correction value and a second correction parameter;

[0064] The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

[0065] In an optional implementation manner, the training module is specifically used for:

[0066] For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter;

[0067] determining a fourth correction value based on a product between the third correction value and a fourth correction parameter;

[0068] The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

[0069] In an optional implementation manner, the training module is specifically used for:

[0070] The parameters of the text error correction model to be trained are adjusted so that the third loss is lower than or equal to the target loss threshold, thereby obtaining a trained text error correction model.

[0071] In a third aspect, an optional implementation of the present disclosure further provides a computer device, a processor, and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory, and when the machine-readable instructions are executed by the processor, the machine-readable instructions perform the steps of the above-mentioned first aspect, or any possible implementation of the first aspect.

[0072] In a fourth aspect, an optional implementation of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are executed.

[0073] For a description of the effects of the above-mentioned text correction device, computer equipment, and computer-readable storage medium, please refer to the description of the above-mentioned text correction method, which will not be repeated here.

[0074] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present disclosure.

[0075] The text correction method, apparatus, computer equipment and storage medium provided by the embodiments of the present disclosure can determine the secondary truth value label corresponding to the primary truth value label of the training sample, and train the text correction model based on the primary truth value label and the secondary truth value label, so that the probability of the secondary truth value label of the training sample determined by the trained text correction model as the true value result is less than the probability of the primary truth value label as the true value result, and greater than the probability of the non-truth value label as the true value result, thereby avoiding the overfitting disadvantage caused by the single correct label during training and reducing false detection.

[0076] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0078] Figure 1 A flowchart of a text error correction method provided by some embodiments of the present disclosure is shown;

[0079] Figure 2 A schematic diagram showing a text error correction device provided by some embodiments of the present disclosure is shown;

[0080] Figure 3 A schematic diagram of a computer device provided by some embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure is not intended to limit the scope of the present disclosure claimed for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0082] Research has found that in the process of training text error correction models, the true value labels of the training data are unique and inconsistent with the actual situation, which makes the model prone to overfitting, resulting in false positives in actual deployment. For example, in the sample sentence "It is inconsistent with the actual feelings of brother", the true value label corresponding to the character "brother" is "condition", but in the actual language environment, there are some other correct answers, such as "situation" and "scenario". Usually, the spelling error correction deep model only sets "condition" as the true value label during training. After training, the model is very likely to correct "situation" and "scenario" as "situation", resulting in false detection and false positives.

[0083] Based on the above research, the present disclosure provides a text correction method, apparatus, computer equipment and storage medium, which can determine the secondary truth value label corresponding to the primary truth value label of the training sample, and train the text correction model based on the primary truth value label and the secondary truth value label, so that the probability of the secondary truth value label of the training sample determined by the trained text correction model as the true value result is less than the probability of the primary truth value label as the true value result, and greater than the probability of the non-true value label as the true value result, thereby avoiding the overfitting disadvantage caused by the single correct label during training and reducing false detection.

[0084] The defects existing in the above solutions are the results obtained by the inventor after practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the present disclosure for the above problems below should be the contributions made by the inventor to the present disclosure during the disclosure process.

[0085] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0086] To facilitate understanding of this embodiment, a text error correction method disclosed in the embodiment of the present disclosure is first introduced in detail. The execution subject of the text error correction method provided in the embodiment of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the text error correction method can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0087] The text error correction method provided by the embodiment of the present disclosure is described below by taking the execution subject as a terminal device as an example.

[0088] See also Figure 1 As shown, it is a flowchart of the text error correction method provided by the embodiment of the present disclosure, and the method includes steps S101 to S104, wherein:

[0089] S101, constructing a text error correction model to be trained, and obtaining multiple training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth label of the incorrect character.

[0090] In this step, a text error correction model can be built. The model can choose traditional machine learning methods, such as Hidden Markov Model, Support Vector Machine (SVM), etc.; or deep learning models can be selected, such as Long Short-Term Memory (LSTM), Transformers model, Bidirectional Encoder Representations from Transformers (BERT), etc.

[0091] Typically, deep learning error correction models can use softmax and cross entropy to calculate losses. loss , and then update the model parameters through reverse gradient propagation to realize the training of deep learning error correction model.

[0092] For example, in the sample sentence "It does not match the actual love brother", assuming that in the dictionary file, the encoding identifier id of "condition" is t The encoding id of "shape" isj , for the error correction model based on deep learning, among the predicted probability values logits (the probability value that the model predicts the correct character corresponding to the error character as the target character) for the character "兄", the score with the id t is . Then the actually calculated loss is:

[0093]

[0094] Among them, i is the i th predicted character corresponding to "兄". Under this loss, the training objective of the model is to make as large as possible, but it does not consider that is also feasible in the sample sentence, resulting in possible error correction of texts such as "情形" to "情况".

[0095] Therefore, the embodiments of the present disclosure propose the concepts of the preferred true value label and the secondary true value label, and train the text error correction model through the preferred true value label and the secondary true value label, so as to avoid overfitting of the model and reduce the probability of false detection.

[0096] Among them, the above-mentioned preferred true value label can be the label that best fits the error character in the training sample. For example, in the training sample "与实际情兄不符", it can be determined that "况" is the preferred true value label corresponding to "兄".

[0097] Exemplarily, the preferred true value label of the error character in the training sample can be determined according to the context information in the training sample, as well as the shape, pronunciation, meaning characteristics, etc. of the error character.

[0098] The above-mentioned training sample can be a piece of text, which can be obtained by transforming the correct text, or by collecting error samples.

[0099] Alternatively, it can also be determined by statistically analyzing a large amount of publicly available text data and according to the number of publicly available data that matches the character segment in the training sample.

[0100] S102. Determine at least one secondary true value label that matches the preferred true value label of the training sample.

[0101] In this step, at least one secondary true value label corresponding to the error character in the training sample can be determined, and the number of secondary true value labels corresponding to each error character can be determined according to the actual situation. Exemplarily, values such as 2, 3, 5, etc. can be taken.

[0102] When determining the secondary truth value label, it can be determined based on the pre-deployed corpus dataset. The corpus dataset can be deployed according to the type of text that needs to be corrected. For example, if Chinese text needs to be corrected, a Chinese corpus dataset can be deployed; if English text needs to be corrected, an English corpus dataset can be deployed; if Chinese and English mixed texts need to be corrected, a Chinese and English mixed corpus dataset can be deployed.

[0103] The above corpus dataset can be generated based on massive amounts of correct text, such as novels, papers, newspapers, etc.

[0104] When determining the secondary truth value label, the erroneous characters in the training sample can be determined first, and the context information of the erroneous characters in the training sample sentence can be determined. According to the context information, a character sequence containing the context information and the pending character is generated. The first character sequence can represent the format of the segment containing the erroneous character in the training sample sentence. The pending character can be a placeholder for the erroneous character in the first character sequence. When searching for a second character sequence matching the first character sequence in the corpus data set, the placeholder can match any character.

[0105] After obtaining the first character sequence, a second character sequence matching the first character sequence can be searched in the corpus data set. When a second character sequence is found, the found second character sequence can be recorded and counted to obtain the corresponding occurrence frequency of the second character sequence.

[0106] Afterwards, a third character sequence may be determined according to the occurrence frequencies of each second character sequence, and the character in the third character sequence that matches the undetermined character may be used as a secondary truth value label.

[0107] Exemplarily, the second character sequence having a frequency higher than a preset frequency may be determined as the third character sequence.

[0108] Specifically, the context information may include the first adjacent character before the position of the erroneous character in the training sample sentence, and the second adjacent character after the position of the erroneous character in the training sample sentence. The number of the first adjacent character and the second adjacent character may be determined according to actual conditions, for example, according to the number of characters in the phrase corresponding to the erroneous character.

[0109] For example, in the training sample sentence "not consistent with the actual brother", the phrase corresponding to the wrong character "brother" is "situation", which has 2 characters, of which the number of wrong characters is 1 and the remaining number of correct characters is 1, then the number of characters between the first adjacent character and the second adjacent character can be 1.

[0110] Alternatively, in the above example, the number of characters of “case” is 2, and then it can be determined that the number of characters of the first character sequence is 3 or 4.

[0111] In one possible implementation, n-grams can be used to extract the first character sequence and the second character sequence. For example, in the above example of "not in line with the actual situation", if the character digits of the first adjacent character and the second adjacent character are 1, then in the way of 3-grams, the first character sequence can be extracted as "qing X bu", where "qing" is the first adjacent character, "bu" is the second adjacent character, and "X" is the undetermined character. The arrangement order of the characters in the first character sequence can be the first adjacent character, the undetermined character, and the second adjacent character.

[0112] When multiple characters are included in the adjacent characters, the order of each character in the adjacent characters can be maintained. For example, if the first adjacent character is "actual situation" and the second adjacent character is "inconsistent", then the first character sequence can be "inconsistent with the actual situation X".

[0113] When the first character sequence is "qing X bu", the second character sequences extracted from the corpus dataset by 3-grams can include "qingkuang bu", "qingxing bu", "qingjing bu", "qingzhi bu", etc.

[0114] When determining the third character sequence, first, from the second character sequence, determine the fourth character sequence whose occurrence frequency is greater than or equal to the preset frequency. Then, according to the occurrence frequency, by sorting the occurrence frequencies of the above-mentioned fourth character sequences, determine the first preset number of fourth character sequences with the most occurrence frequencies as the third character sequence.

[0115] For example, if the occurrence frequencies corresponding to the second character sequences "qingkuang bu", "qingxing bu", "qingzhi bu", "qingjing bu" are 1506021, 1305055, 1131768, 983579 respectively, the preset frequency is 1000000, and the preset number is 2, and "qingkuang bu" is the second character sequence corresponding to the preferred true label, then first remove "qingkuang bu" from the second character sequence, and remove "qingjing bu" with an occurrence frequency lower than 1000000 from the second character sequence, to obtain the fourth character sequences "qingxing bu" and "qingzhi bu". Then, determine the first 2 "qingxing bu" and "qingzhi bu" with the most occurrence frequencies as the third character sequence.

[0116] S103. Use the training sample carrying the preferred true label and the secondary preferred true label corresponding to the training sample to train the text error correction model to be trained, and obtain a trained text error correction model; the trained text error correction model makes the probability that the secondary preferred true label of the training sample is used as the true value result less than the probability that the preferred true label is used as the true value result, and greater than the probability that the non-true label is used as the true value result.

[0117] In this step, the training samples can be first input into the text correction model to be trained. Through the text correction model, the prediction results of multiple target characters corresponding to the erroneous characters can be determined. The prediction results can indicate the probability that the correct character corresponding to the erroneous character is the target character.

[0118] When the above text error correction model is working, it can first determine multiple target characters, then determine the probability that the correct character corresponding to the erroneous character is the above target characters, perform error correction based on the determined probability, and determine the correct character corresponding to the erroneous character.

[0119] After obtaining the prediction results of multiple target characters corresponding to the erroneous characters, the probability corresponding to the secondary truth label (the above-mentioned target character includes the secondary truth label) can be determined from the prediction results, and the first loss can be determined based on the first difference between the probability corresponding to the secondary truth label and the probability corresponding to the primary truth label.

[0120] At the same time, the second difference between the probabilities corresponding to other target characters in the prediction results except the second-choice truth label and the first-choice truth label and the probability corresponding to the second-choice truth label can be determined, and the second loss can be determined based on the second difference.

[0121] Afterwards, the third loss may be determined based on the sum of the first loss and the second loss.

[0122] Finally, the third loss can be used to train the text error correction model to be trained to obtain a trained text error correction model.

[0123] When using the third loss for training, the training goal can be to minimize the value of the third loss. Since the third loss is determined based on the sum of the first loss and the second loss, the training goal is equivalent to minimizing the first loss and the second loss and minimizing the sum of the two; that is, minimizing the first difference and the second difference.

[0124] In this way, when the trained text error correction model is working, the probability of the secondary truth value label of the training sample being the true value result is less than the probability of the primary truth value label being the true value result, and greater than the probability of the non-true value label being the true value result. The primary truth value label will be prioritized as the true value result, and the importance of the secondary truth value label will be higher than that of other non-true value labels, which can avoid overfitting and thus reduce the probability of false detection.

[0125] Specifically, the first loss can be determined according to the following steps:

[0126] For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter;

[0127] determining a second correction value based on a product of the first correction value and a second correction parameter;

[0128] The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

[0129] In this step, a first correction parameter and a second correction parameter may be introduced to correct the first difference. The first correction parameter may scale the size of the first difference, thereby adjusting the difference in importance between the first true value label and the second true value label; the second correction parameter may amplify or reduce the first difference, thereby making the first loss closer to the actual required state.

[0130] Accordingly, a third correction parameter and a fourth correction parameter may also be introduced, and the second loss may be determined by the following steps:

[0131] For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter;

[0132] determining a fourth correction value based on a product between the third correction value and a fourth correction parameter;

[0133] The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

[0134] In this step, the third correction parameter can scale the size of the second difference, thereby adjusting the difference in importance between the secondary true value label and other target characters; the fourth correction parameter can enlarge or reduce the second difference, thereby making the second loss closer to the actual required state.

[0135] Exemplarily, the third loss may be determined by the following formula:

[0136]

[0137] in, loss For the third loss, No. j The probability corresponding to the second-choice true value label, is the probability corresponding to the preferred true value label, is the first correction value, is the second correction value, is the set corresponding to the second-choice truth label, is a set of other target characters, For the i The probability corresponding to other target characters, is the third correction value, This is the fourth correction value.

[0138] The values ​​of the first correction value, the second correction value, the third correction value and the fourth correction value can be determined according to actual conditions. For example, the first correction value can be 0.01, the second correction value can be 0.9, the third correction value can be 0.02, and the fourth correction value can be 1.1.

[0139] The learning goal of the third loss can be to make the score (i.e., probability) of the first true value label greater than the score of the second true value label, and the score of the second true value label greater than the scores of other incorrect labels (i.e., other target labels). At the same time, the importance of different learning goals during model learning is adjusted through hyperparameters (i.e., the first correction value, the second correction value, the third correction value, and the fourth correction value).

[0140] During the training process, the third loss can be reduced by adjusting the model parameters. When the third loss is reduced to a certain level (for example, lower than a preset threshold), or when the third loss converges (the third loss remains basically unchanged after adjusting the parameters), the training can be stopped to obtain a trained text error correction model.

[0141] S104, obtaining the text to be corrected, inputting the text to be corrected into the trained text correction model, and obtaining the correction result of the text to be corrected.

[0142] After completing the training of the text correction model, the trained text correction model can be used to correct the text. The text to be corrected is input into the trained text correction model, and the correction result output by the text correction model can be obtained. The correction result may include the corrected text, and / or the correct characters corresponding to the erroneous characters in the text to be corrected.

[0143] In this way, the trained text correction model can minimize the probability of overfitting and increase the importance of the secondary true value label. In the text to be corrected "not consistent with the actual situation", "brother" can be correctly corrected to "situation", and in the text "not consistent with the actual situation", since the importance of "shape" in "situation" is increased during the training process, "shape" will not be corrected to "situation", reducing the probability of false detection.

[0144] The text correction method provided by the embodiment of the present disclosure can determine the secondary truth value label corresponding to the primary truth value label of the training sample, and train the text correction model based on the primary truth value label and the secondary truth value label, so that the probability of the secondary truth value label of the training sample determined by the trained text correction model as the true value result is smaller than the probability of the primary truth value label as the true value result, and larger than the probability of the non-truth value label as the true value result, thereby avoiding the overfitting disadvantage caused by the single correct label during training and reducing false detection.

[0145] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0146] Based on the same inventive concept, a text correction device corresponding to the text correction method is also provided in the embodiment of the present disclosure. Since the principle of solving the problem by the device in the embodiment of the present disclosure is similar to the above-mentioned text correction method in the embodiment of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0147] Reference Figure 2 FIG. 1 is a schematic diagram of a text error correction device provided by an embodiment of the present disclosure, wherein the device comprises:

[0148] A construction module 210 is used to construct a text error correction model to be trained, and obtain multiple training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character;

[0149] A determination module 220, configured to determine at least one secondary truth label that matches the primary truth label of the training sample;

[0150] The training module 230 is used to train the text error correction model to be trained using the training sample carrying the first truth value label and the second truth value label corresponding to the training sample to obtain a trained text error correction model; the trained text error correction model makes the probability of the second truth value label of the training sample being the truth value result less than the probability of the first truth value label being the truth value result, and greater than the probability of the non-truth value label being the truth value result;

[0151] The error correction module 240 is used to obtain the text to be corrected, input the text to be corrected into the trained text error correction model, and obtain the error correction result of the text to be corrected.

[0152] The text correction device provided by the embodiment of the present disclosure can determine the secondary truth value label corresponding to the primary truth value label of the training sample, and train the text correction model based on the primary truth value label and the secondary truth value label, so that the probability of the secondary truth value label of the training sample determined by the trained text correction model as the true value result is smaller than the probability of the primary truth value label as the true value result, and greater than the probability of the non-truth value label as the true value result, thereby avoiding the overfitting disadvantage caused by the single correct label during training and reducing false detection.

[0153] In an optional implementation manner, the determining module 220 is specifically configured to:

[0154] For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence;

[0155] Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence;

[0156] Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies;

[0157] Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

[0158] In an optional implementation manner, the determining module 220 is specifically configured to:

[0159] Determine a first adjacent character before the position where the erroneous character is located in the training sample sentence, and a second adjacent character after the position where the erroneous character is located in the training sample sentence;

[0160] A first character sequence is generated based on the first adjacent character and the second adjacent character; the first character sequence includes the first adjacent character, the to-be-determined character, and the second adjacent character arranged in sequence.

[0161] In an optional implementation manner, the determining module 220 is specifically configured to:

[0162] Determine, from the second character sequence, a fourth character sequence whose occurrence frequency is greater than or equal to a preset frequency;

[0163] sorting the fourth character sequence according to the occurrence frequency;

[0164] The fourth character sequences of the first preset number with the highest occurrence frequency are determined as the third character sequence.

[0165] In an optional implementation manner, the training module 230 is specifically used for:

[0166] Inputting the training sample into the text error correction model to be trained, obtaining prediction results of multiple target characters corresponding to the erroneous character determined by the text error correction model to be trained; the prediction results represent the probability that the correct character corresponding to the erroneous character is the target character;

[0167] Determine a first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label;

[0168] Determine a second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label;

[0169] determining a third loss according to the sum of the first loss and the second loss;

[0170] The text error correction model to be trained is trained using the third loss to obtain a trained text error correction model.

[0171] In an optional implementation manner, the training module 230 is specifically used for:

[0172] For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter;

[0173] determining a second correction value based on a product of the first correction value and a second correction parameter;

[0174] The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

[0175] In an optional implementation manner, the training module 230 is specifically used for:

[0176] For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter;

[0177] determining a fourth correction value based on a product between the third correction value and a fourth correction parameter;

[0178] The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

[0179] In an optional implementation manner, the training module 230 is specifically used for:

[0180] The parameters of the text error correction model to be trained are adjusted so that the third loss is lower than or equal to the target loss threshold, thereby obtaining a trained text error correction model.

[0181] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0182] The present disclosure also provides a computer device, such as Figure 3 FIG. 1 is a schematic diagram of a computer device structure provided in an embodiment of the present disclosure, including:

[0183] A processor 31 and a memory 32; the memory 32 stores machine-readable instructions executable by the processor 31, and the processor 31 is used to execute the machine-readable instructions stored in the memory 32. When the machine-readable instructions are executed by the processor 31, the processor 31 performs the following steps:

[0184] Constructing a text error correction model to be trained, and obtaining a plurality of training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character;

[0185] Determining at least one secondary truth label that matches the primary truth label of the training sample;

[0186] The text error correction model to be trained is trained using the training sample carrying the preferred true value label and the secondary true value label corresponding to the training sample to obtain a trained text error correction model; the trained text error correction model makes the probability that the secondary true value label of the training sample is the true value result less than the probability that the preferred true value label is the true value result, and greater than the probability that the non-true value label is the true value result;

[0187] The text to be corrected is obtained, and the text to be corrected is input into the trained text correction model to obtain the correction result of the text to be corrected.

[0188] In an optional implementation, in the instructions executed by the processor 31, the determining of at least one secondary truth label matching the primary truth label of the training sample includes:

[0189] For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence;

[0190] Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence;

[0191] Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies;

[0192] Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

[0193] In an optional implementation manner, in the instructions executed by the processor 31, generating a first character sequence containing the context information and the undetermined character based on the context information of the erroneous character in the training sample sentence includes:

[0194] Determine a first adjacent character before the position where the erroneous character is located in the training sample sentence, and a second adjacent character after the position where the erroneous character is located in the training sample sentence;

[0195] A first character sequence is generated based on the first adjacent character and the second adjacent character; the first character sequence includes the first adjacent character, the to-be-determined character, and the second adjacent character arranged in sequence.

[0196] In an optional implementation manner, in the instructions executed by the processor 31, determining the third character sequence from the multiple second character sequences according to the occurrence frequency includes:

[0197] Determine, from the second character sequence, a fourth character sequence whose occurrence frequency is greater than or equal to a preset frequency;

[0198] sorting the fourth character sequence according to the occurrence frequency;

[0199] The fourth character sequences of the first preset number with the highest occurrence frequency are determined as the third character sequence.

[0200] In an optional implementation, in the instructions executed by the processor 31, the training sample carrying the preferred truth value label and the secondary truth value label corresponding to the training sample are used to train the text error correction model to be trained to obtain the trained text error correction model, including:

[0201] Inputting the training sample into the text error correction model to be trained, obtaining prediction results of multiple target characters corresponding to the erroneous character determined by the text error correction model to be trained; the prediction results represent the probability that the correct character corresponding to the erroneous character is the target character;

[0202] Determine a first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label;

[0203] Determine a second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label;

[0204] determining a third loss according to the sum of the first loss and the second loss;

[0205] The text error correction model to be trained is trained using the third loss to obtain a trained text error correction model.

[0206] In an optional implementation, in the instructions executed by the processor 31, determining the first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label includes:

[0207] For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter;

[0208] determining a second correction value based on a product of the first correction value and a second correction parameter;

[0209] The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

[0210] In an optional implementation, in the instructions executed by the processor 31, determining the second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label includes:

[0211] For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter;

[0212] determining a fourth correction value based on a product between the third correction value and a fourth correction parameter;

[0213] The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

[0214] In an optional implementation manner, in the instructions executed by the processor 31, the step of training the text error correction model to be trained by using the third loss to obtain the trained text error correction model includes:

[0215] The parameters of the text error correction model to be trained are adjusted so that the third loss is lower than or equal to the target loss threshold, thereby obtaining a trained text error correction model.

[0216] The above-mentioned memory 32 includes internal memory 321 and external memory 322; the memory 321 here is also called internal memory, which is used to temporarily store the calculation data in the processor 31, as well as the data exchanged with the external memory 322 such as the hard disk. The processor 31 exchanges data with the external memory 322 through the internal memory 321.

[0217] The specific execution process of the above instructions can refer to the steps of the text error correction method described in the embodiment of the present disclosure, which will not be repeated here.

[0218] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the text error correction method described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0219] The present disclosure also provides a computer program product that carries a program code. The program code includes instructions that can be used to execute the steps of the text error correction method described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0220] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0221] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0222] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0223] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0224] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0225] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be based on the protection scope of the claims.

Claims

1. A text error correction method, characterized in that: include: Constructing a text error correction model to be trained, and obtaining a plurality of training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character; Determining at least one secondary truth label that matches the primary truth label of the training sample; The text error correction model to be trained is trained using the training sample carrying the preferred true value label and the secondary true value label corresponding to the training sample to obtain a trained text error correction model; the trained text error correction model makes the probability that the secondary true value label of the training sample is the true value result less than the probability that the preferred true value label is the true value result, and greater than the probability that the non-true value label is the true value result; Obtaining a text to be corrected, inputting the text to be corrected into the trained text correction model, and obtaining a correction result of the text to be corrected; The determining of at least one secondary truth label that matches the primary truth label of the training sample comprises: For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence; Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence; Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies; Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

2. The method according to claim 1, characterized in that: The step of generating a first character sequence containing the context information and a pending character based on the context information of the erroneous character in the training sample sentence includes: Determine a first adjacent character before the position where the erroneous character is located in the training sample sentence, and a second adjacent character after the position where the erroneous character is located in the training sample sentence; A first character sequence is generated based on the first adjacent character and the second adjacent character; the first character sequence includes the first adjacent character, the to-be-determined character, and the second adjacent character arranged in sequence.

3. The method according to claim 1, characterized in that: Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies includes: Determine, from the second character sequence, a fourth character sequence whose occurrence frequency is greater than or equal to a preset frequency; sorting the fourth character sequence according to the occurrence frequency; The fourth character sequences of the first preset number with the highest occurrence frequency are determined as the third character sequence.

4. The method according to claim 1, characterized in that: The method of training the text error correction model to be trained by using the training sample carrying the preferred true value label and the secondary true value label corresponding to the training sample to obtain the trained text error correction model includes: Inputting the training sample into the text error correction model to be trained, and obtaining prediction results of multiple target characters corresponding to the erroneous character determined by the text error correction model to be trained; the prediction results represent the probability that the correct character corresponding to the erroneous character is the target character; Determine a first loss according to a first difference between a probability corresponding to the second selected true value label in the prediction result and a probability corresponding to the first selected true value label; Determine a second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label; determining a third loss according to the sum of the first loss and the second loss; The text error correction model to be trained is trained using the third loss to obtain a trained text error correction model.

5. The method according to claim 4, characterized in that The determining the first loss according to a first difference between the probability corresponding to the second selected true value label in the prediction result and the probability corresponding to the first selected true value label includes: For each probability corresponding to the second selected truth value label, determine a first correction value of the probability corresponding to the second selected truth value label based on the first difference corresponding to the probability corresponding to the second selected truth value label and the sum of the first correction parameter; determining a second correction value based on a product of the first correction value and a second correction parameter; The first loss is determined based on the sum of the second corrected values ​​of the probabilities corresponding to the second-selected true value labels.

6. The method according to claim 4, characterized in that The determining the second loss according to a second difference between the probabilities corresponding to other target characters in the prediction result except the second selected true value label and the first selected true value label and the probability corresponding to the second selected true value label comprises: For each probability corresponding to the other target character, determine a third correction value corresponding to the other target character based on the second difference corresponding to the probability corresponding to the other target character and the sum of the third correction parameter; determining a fourth correction value based on a product between the third correction value and a fourth correction parameter; The second loss is determined based on the sum of the fourth correction values ​​of the probabilities corresponding to the other target characters.

7. The method according to claim 4, characterized in that The method of training the text error correction model to be trained by using the third loss to obtain a trained text error correction model includes: The parameters of the text error correction model to be trained are adjusted so that the third loss is lower than or equal to the target loss threshold, thereby obtaining a trained text error correction model.

8. A text error correction device, characterized in that: include: A construction module is used to construct a text error correction model to be trained, and obtain multiple training samples of the text error correction model, wherein the training samples are sentences including at least one incorrect character, and the correct character corresponding to the incorrect character is the preferred truth value label of the incorrect character; A determination module, configured to determine at least one secondary truth label that matches the primary truth label of the training sample; A training module, used to train the text error correction model to be trained using the training samples carrying the preferred true value labels and the secondary true value labels corresponding to the training samples, to obtain a trained text error correction model; the trained text error correction model makes the probability of the secondary true value label of the training sample being the true value result less than the probability of the preferred true value label being the true value result, and greater than the probability of the non-true value label being the true value result; The error correction module is used to obtain the text to be corrected, input the text to be corrected into the trained text error correction model, and obtain the error correction result of the text to be corrected; The determination module is specifically used for: For any erroneous character, based on context information of the erroneous character in the training sample sentence, generate a first character sequence containing the context information and a pending character; the pending character is a placeholder for the erroneous character in the first character sequence; Searching for multiple second character sequences matching the first character sequence from the corpus data set, and determining the occurrence frequency corresponding to each second character sequence; Determining a third character sequence from the plurality of second character sequences according to the occurrence frequencies; Determine in the third character sequence that the character that matches the pending character is the secondary selected truth value label.

9. A computer device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor executes the steps of the text error correction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program. When the computer program is executed by a computer device, the computer device executes the steps of the text error correction method according to any one of claims 1 to 7.

Images