Target-level sentence simplification method and system based on label confidence coefficient weighting

Abstract

The invention discloses a target-level sentence simplification method and system based on label confidence coefficient weighting, which can capture nuisance of sentences on different simplification levels by training a multi-level simplification classifier, and the multi-level classification not only improves the classification accuracy, but also enables the subsequent sentence simplification process to be more meticulous and flexible. The source sentences and the target sentences of the paraphrasing data set are marked by utilizing the multi-stage simplified classifier, a large amount of pseudo training data can be automatically generated, the problem of scarcity of high-quality marked data can be effectively relieved, the cost of data collection is reduced, and when a sentence simplification model is trained, the training efficiency is improved. By introducing a label confidence coefficient weighting scheme, the influence of error label samples on model training can be remarkably reduced, the robustness and the performance stability of the model can be improved, the finally obtained target-level sentence simplification model can receive a sentence to be simplified as input and output the simplified sentence, and the sentence simplification efficiency is improved. The output simplified sentences are generally high in quality and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of sentence simplification, and in particular to a target-level sentence simplification method and system based on label confidence weighting. Background Art

[0002] In the field of natural language processing, sentence simplification is an important research direction, which aims to convert complex sentences into simpler and more understandable forms while maintaining the main information and meaning of the original sentence. Traditional sentence simplification methods often rely on complex rule systems or large-scale corpora for statistical learning, but these methods often face challenges in dealing with diverse and complex sentences.

[0003] In recent years, with the rapid development of deep learning technology, sentence simplification models based on neural networks have gradually demonstrated powerful capabilities. These models can learn the complex patterns and rules of sentence simplification by training on large-scale data sets. However, high-quality data sets with simplification level labels are usually scarce, which limits the further improvement of model performance and results in poor stability and accuracy of sentence simplification effects. Summary of the invention

[0004] In view of this, the present invention proposes a target-level sentence simplification method and system based on label confidence weighting, which can solve the defects of poor stability and accuracy of sentence simplification effect in the prior art.

[0005] The technical solution of the present invention is achieved in this way:

[0006] A target-level sentence simplification method based on label confidence weighting, specifically comprising:

[0007] Train a multi-level simplified classifier based on the target-level benchmark dataset;

[0008] The source and target sentences of the paraphrase dataset are labeled according to the multi-level simplified classifier to obtain pseudo training data;

[0009] Train based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model;

[0010] The sentence to be simplified is input into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

[0011] As a further optional solution of the target-level sentence simplification method based on label confidence weighting, the multi-level simplified classifier is trained based on the target-level benchmark dataset, specifically including:

[0012] Obtaining object-level benchmark datasets from a multi-level text simplification corpus;

[0013] Select the BERT model as the training model and modify the output layer of the BERT model so that the number of neurons in its output layer matches the number of sentence simplification levels;

[0014] Input the target-level benchmark dataset into the modified BERT model to obtain the predicted simplified level probability distribution;

[0015] The loss between the probability distribution predicted by the model and the true label is calculated according to the cross entropy loss function to obtain the loss value;

[0016] Based on the loss value, use the optimizer to perform backpropagation, calculate the gradient and update the model parameters;

[0017] Repeat the above steps until the preset training rounds are reached to obtain a multi-level simplified classifier.

[0018] As a further optional solution of the target-level sentence simplification method based on label confidence weighting, the source sentences and target sentences of the interpretation data set are labeled according to the multi-level simplified classifier to obtain pseudo training data, which specifically includes:

[0019] Acquire a paraphrase dataset, wherein the paraphrase dataset includes multiple pairs of source sentences x and target sentences x';

[0020] For each pair of sentences (x, x') in the paraphrase dataset, use a multi-level simplification classifier to predict the source sentence x and the target sentence x' respectively, and obtain their respective simplification levels y and y', as well as the sentence confidence scores c and c' for these two levels;

[0021] For each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for these two levels are recorded to form pseudo training data.

[0022] As a further optional solution of the target-level sentence simplification method based on label confidence weighting, the training based on label confidence weighting and pseudo training data to obtain the target-level sentence simplification model specifically includes:

[0023] Calculate the accuracy p(y) of the multi-level simplified classifier for each simplification level y;

[0024] Get the sentence confidence score c for each simplification level y;

[0025] Perform a weighted label confidence calculation based on the accuracy p(y) and sentence confidence score c of each simplification level y to obtain the label confidence score w;

[0026] The sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w are input into the BART model for training to obtain the target-level sentence simplification model.

[0027] As a further optional scheme of the target-level sentence simplification method based on label confidence weighting, in the process of inputting the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training, a weighted cross entropy loss function is used to weight the loss of each sample according to the label confidence score w of each sample, thereby adjusting the training loss.

[0028] A target-level sentence simplification system based on label confidence weighting, comprising:

[0029] A first training module is used to train a multi-level simplified classifier based on a target-level benchmark data set;

[0030] A labeling module is used to label the source sentences and target sentences of the interpretation dataset according to a multi-level simplified classifier to obtain pseudo training data;

[0031] The second training module is used to perform training based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model;

[0032] The input module is used to input the sentence to be simplified into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

[0033] As a further optional solution of the target-level sentence simplification system based on label confidence weighting, the tagging module includes:

[0034] A first acquisition module is used to acquire a paraphrase dataset, wherein the paraphrase dataset includes a plurality of pairs of source sentences x and target sentences x';

[0035] The prediction module is used to predict the source sentence x and the target sentence x' for each pair of sentences (x, x') in the paraphrase dataset using a multi-level simplification classifier, and obtain the respective simplification levels y and y', as well as the sentence confidence scores c and c' for the two levels;

[0036] The recording module is used to record, for each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for the two levels of sentences, to form pseudo training data.

[0037] As a further optional solution of the target-level sentence simplification system based on label confidence weighting, the second training module includes:

[0038] An accuracy calculation module, used to calculate the accuracy p(y) of the multi-level simplified classifier for each simplified level y;

[0039] A second acquisition module is used to obtain a sentence confidence score c for each simplification level y;

[0040] A label confidence weighted calculation module is used to perform label confidence weighted calculation according to the accuracy p(y) of each simplification level y and the sentence confidence score c to obtain a label confidence score w;

[0041] The execution module is used to input the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training to obtain a target-level sentence simplification model.

[0042] A computing device comprises 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 steps of any one of the above-mentioned target-level sentence simplification methods based on label confidence weighting are implemented.

[0043] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any of the steps of the above-mentioned target-level sentence simplification method based on label confidence weighting.

[0044] The beneficial effects of the present invention are as follows: by training a multi-level simplified classifier, the subtle differences of sentences at different simplification levels can be captured. This multi-level classification not only improves the accuracy of classification, but also makes the subsequent sentence simplification process more detailed and flexible. Different levels of simplification can provide more personalized solutions for different types of sentences and needs. By using a multi-level simplified classifier to mark the source sentences and target sentences of the interpretation data set, a large amount of pseudo training data can be automatically generated, which can effectively alleviate the problem of scarcity of high-quality annotated data and reduce the cost of data collection. At the same time, the diversity and richness of the pseudo training data also help to improve the generalization ability of the sentence simplification model. When training the sentence simplification model, introducing a label confidence weighting scheme can significantly reduce the impact of incorrectly labeled samples on model training. By assigning a weight proportional to the confidence of its label to each sample, the model can pay more attention to samples with accurate labeling and high confidence, thereby learning more accurate and effective simplification rules. This strategy helps to improve the robustness and performance stability of the model. The final target-level sentence simplification model can receive the sentence to be simplified as input and output the simplified sentence. Since the model fully considers the simplification level and label confidence of the sentence during the training process, the simplified sentences it outputs are usually of high quality and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A flowchart of a target-level sentence simplification method based on label confidence weighting according to the present invention;

[0047] Figure 2 The present invention is a schematic diagram of the composition of a target-level sentence simplification system based on label confidence weighting. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] refer to Figure 1-2 , a target-level sentence simplification method based on label confidence weighting, specifically including:

[0050] Train a multi-level simplified classifier based on the target-level benchmark dataset;

[0051] The source and target sentences of the paraphrase dataset are labeled according to the multi-level simplified classifier to obtain pseudo training data;

[0052] Train based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model;

[0053] The sentence to be simplified is input into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

[0054] In this embodiment, by training a multi-level simplified classifier, the subtle differences of sentences at different simplification levels can be captured. This multi-level classification not only improves the accuracy of classification, but also makes the subsequent sentence simplification process more detailed and flexible. Different levels of simplification can provide more personalized solutions for different types of sentences and needs. By using a multi-level simplified classifier to mark the source sentences and target sentences of the interpretation dataset, a large amount of pseudo training data can be automatically generated, which can effectively alleviate the problem of scarcity of high-quality labeled data and reduce the cost of data collection. At the same time, the diversity and richness of the pseudo training data also help to improve the generalization ability of the sentence simplification model. When training the sentence simplification model, introducing a label confidence weighting scheme can significantly reduce the impact of incorrectly labeled samples on model training. By assigning a weight proportional to the confidence of its label to each sample, the model can pay more attention to samples with accurate labeling and high confidence, thereby learning more accurate and effective simplification rules. This strategy helps to improve the robustness and performance stability of the model. The final target-level sentence simplification model can receive the sentence to be simplified as input and output the simplified sentence. Since the model fully considers the simplification level and label confidence of the sentence during the training process, the simplified sentences it outputs are usually of higher quality and accuracy.

[0055] Preferably, the training of a multi-level simplified classifier based on the target-level benchmark data set specifically includes:

[0056] Obtaining object-level benchmark datasets from a multi-level text simplification corpus;

[0057] Select the BERT model as the training model and modify the output layer of the BERT model so that the number of neurons in its output layer matches the number of sentence simplification levels;

[0058] Input the target-level benchmark dataset into the modified BERT model to obtain the predicted simplified level probability distribution;

[0059] The loss between the probability distribution predicted by the model and the true label is calculated according to the cross entropy loss function to obtain the loss value;

[0060] Based on the loss value, use the optimizer to perform backpropagation, calculate the gradient and update the model parameters;

[0061] Repeat the above steps until the preset training rounds are reached to obtain a multi-level simplified classifier.

[0062] In this embodiment, the BERT model, as a pre-trained language model, has a strong semantic representation capability. By modifying its output layer to match the number of sentence simplification levels, the model can be adapted to a specific simplification task. This design enables the model to more accurately capture the association between sentences and simplification levels, thereby improving the accuracy of classification. During the training process, the model can output the probability distribution of each sentence belonging to different simplification levels by processing the input target-level benchmark data set. This probability distribution not only provides the prediction result of the classification, but also reflects the confidence of the model in different simplification levels, providing a basis for subsequent training adjustments. The cross entropy loss function is an effective method to measure the difference between the probability distribution predicted by the model and the true label. By calculating the loss value, the performance of the model during the training process can be quantified and used as the basis for the optimizer to adjust the model parameters. The selection of this loss function helps the model to converge quickly to the optimal solution. After obtaining the loss value, the optimizer calculates the gradient through the back-propagation algorithm and updates the model parameters according to the gradient. This process is continuously iterated, so that the model can gradually learn more accurate classification rules. The use of the optimizer improves the efficiency and stability of the training process, which helps the model achieve a higher performance level within a limited time.

[0063] Preferably, the step of labeling the source sentences and the target sentences of the interpretation data set according to the multi-level simplified classifier to obtain pseudo training data specifically includes:

[0064] Acquire a paraphrase dataset, wherein the paraphrase dataset includes multiple pairs of source sentences x and target sentences x';

[0065] For each pair of sentences (x, x') in the paraphrase dataset, use a multi-level simplification classifier to predict the source sentence x and the target sentence x' respectively, and obtain their respective simplification levels y and y', as well as the sentence confidence scores c and c' for these two levels;

[0066] For each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for these two levels are recorded to form pseudo training data.

[0067] In this embodiment, the source and target sentences in the interpretation data set are predicted by a multi-level simplified classifier, and their corresponding simplification levels (y and y') and corresponding confidence scores (c and c') can be obtained. This step fully utilizes the advantages of the multi-level classifier and can more accurately capture the complexity and diversity of sentences in the simplification process. The introduction of the sentence confidence score further quantifies the classifier's grasp of the prediction results, providing an important reference for subsequent data processing and analysis; the predicted simplification level (y and y') and confidence score (c and c') of each pair of sentences (x, x') are recorded to form pseudo training data. This process Not only does it retain the information of the original sentence pairs, but it also adds additional information about their simplification levels and confidences, which is of great significance for the subsequent training of more complex and sophisticated sentence simplification models. The pseudo training data generated by the above steps not only expands the scale and diversity of the training data, but also improves the label quality of the data. Since these pseudo training data are generated based on the prediction results of the multi-level simplified classifier, they reflect the knowledge and experience of the classifier to a certain extent. This method of generating pseudo training data based on prediction results can help alleviate the problem of scarcity of high-quality labeled data, and also help improve the generalization ability and robustness of the sentence simplification model.

[0068] Preferably, the training based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model specifically includes:

[0069] Calculate the accuracy p(y) of the multi-level simplified classifier for each simplification level y;

[0070] Get the sentence confidence score c for each simplification level y;

[0071] Perform a weighted label confidence calculation based on the accuracy p(y) and sentence confidence score c of each simplification level y to obtain the label confidence score w;

[0072] The sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w are input into the BART model for training to obtain the target-level sentence simplification model.

[0073] In this embodiment, precision is one of the important indicators for measuring classifier performance. It reflects the accuracy of the classifier at a specific simplification level. By calculating the precision of the multi-level simplified classifier at each simplification level, the performance of the classifier at different simplification difficulties can be understood. This step provides an important reference for the subsequent label confidence weighted calculation; the sentence confidence score is a quantitative representation of the reliability of the classifier's prediction results. In the prediction process of the multi-level simplified classifier, each simplification level will correspond to a confidence score, reflecting the degree of the classifier's grasp of the prediction results at that level. Obtaining these confidence scores helps to more accurately evaluate the reliability of the labels in the pseudo training data; by combining the precision p(y) and the sentence confidence score c, each label can be calculated. The confidence score w is calculated by this step. This step fully considers the performance differences of the classifier at different simplification levels and the reliability of the prediction results, and assigns a more reasonable weight to each label in the pseudo training data. This weighting method helps to reduce the impact of incorrect labels on model training and improve the quality of training data and the effect of model training. As a powerful text generation model, the BART model is suitable for tasks such as sentence simplification. By inputting the weighted pseudo training data into the BART model for training, the rich information and label confidence scores in the pseudo training data can be fully utilized, so that the model can learn more accurate and effective simplification rules. This training method not only improves the generalization ability of the model, but also enables the model to pay more attention to the preservation of semantics and the improvement of fluency when simplifying sentences.

[0074] Preferably, in the process of inputting the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training, a weighted cross entropy loss function is used to weight the loss of each sample according to the label confidence score w of each sample, thereby adjusting the training loss.

[0075] In this embodiment, during the training process, the weighted cross entropy loss function adjusts the contribution of each sample to the total loss according to its weight, which means that the model will pay more attention to samples that have a greater impact on the optimization target, thereby accelerating the convergence speed of the model and improving the training efficiency. In addition, by dynamically adjusting the weights, the model can more flexibly adapt to training scenarios with different data distributions and complexities; the model trained using the weighted cross entropy loss function usually has improved performance, because the model has learned how to more accurately process samples of different confidence levels during the training process, so when faced with unknown or complex inputs, the model can make more reasonable predictions and decisions. This performance improvement is not only reflected in conventional evaluation indicators such as accuracy and recall, but may also be reflected in higher-level capabilities such as the robustness and generalization ability of the model; through the training method of the weighted cross entropy loss function, the model can pay more attention to samples that can accurately reflect the semantics of the original sentence, which helps the model to simplify the sentence while maintaining its original meaning and information content, thereby enhancing the model's semantic understanding ability.

[0076] A target-level sentence simplification system based on label confidence weighting, comprising:

[0077] A first training module is used to train a multi-level simplified classifier based on a target-level benchmark data set;

[0078] A labeling module is used to label the source sentences and target sentences of the interpretation dataset according to a multi-level simplified classifier to obtain pseudo training data;

[0079] The second training module is used to perform training based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model;

[0080] The input module is used to input the sentence to be simplified into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

[0081] Preferably, the marking module comprises:

[0082] A first acquisition module is used to acquire a paraphrase dataset, wherein the paraphrase dataset includes a plurality of pairs of source sentences x and target sentences x';

[0083] The prediction module is used to predict the source sentence x and the target sentence x' for each pair of sentences (x, x') in the paraphrase dataset using a multi-level simplification classifier, and obtain the respective simplification levels y and y', as well as the sentence confidence scores c and c' for the two levels;

[0084] The recording module is used to record, for each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for the two levels of sentences, to form pseudo training data.

[0085] Preferably, the second training module includes:

[0086] An accuracy calculation module, used to calculate the accuracy p(y) of the multi-level simplified classifier for each simplified level y;

[0087] A second acquisition module is used to obtain a sentence confidence score c for each simplification level y;

[0088] A label confidence weighted calculation module is used to perform label confidence weighted calculation according to the accuracy p(y) of each simplification level y and the sentence confidence score c to obtain a label confidence score w;

[0089] The execution module is used to input the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training to obtain a target-level sentence simplification model.

[0090] A computing device comprises 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 steps of any one of the above-mentioned target-level sentence simplification methods based on label confidence weighting are implemented.

[0091] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any of the steps of the above-mentioned target-level sentence simplification method based on label confidence weighting.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A target-level sentence simplification method based on label confidence weighting, characterized in that: Specifically include: Train a multi-level simplified classifier based on the target-level benchmark dataset; The source and target sentences of the paraphrase dataset are labeled according to the multi-level simplified classifier to obtain pseudo training data; Train based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model; The sentence to be simplified is input into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

2. The target-level sentence simplification method based on label confidence weighting according to claim 1, characterized in that: The method of training a multi-level simplified classifier based on the target-level benchmark dataset specifically includes: Obtaining object-level benchmark datasets from a multi-level text simplification corpus; Select the BERT model as the training model and modify the output layer of the BERT model so that the number of neurons in its output layer matches the number of sentence simplification levels; Input the target-level benchmark dataset into the modified BERT model to obtain the predicted simplified level probability distribution; The loss between the probability distribution predicted by the model and the true label is calculated according to the cross entropy loss function to obtain the loss value; Based on the loss value, use the optimizer to perform backpropagation, calculate the gradient and update the model parameters; Repeat the above steps until the preset training rounds are reached to obtain a multi-level simplified classifier.

3. The target-level sentence simplification method based on label confidence weighting according to claim 2, characterized in that: The method of labeling the source sentences and target sentences of the interpretation data set according to the multi-level simplified classifier to obtain pseudo training data specifically includes: Acquire a paraphrase dataset, wherein the paraphrase dataset includes multiple pairs of source sentences x and target sentences x'; For each pair of sentences (x, x') in the paraphrase dataset, use a multi-level simplification classifier to predict the source sentence x and the target sentence x' respectively, and obtain their respective simplification levels y and y', as well as the sentence confidence scores c and c' for these two levels; For each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for these two levels are recorded to form pseudo training data.

4. The target-level sentence simplification method based on label confidence weighting according to claim 3, characterized in that: The training is performed based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model, which specifically includes: Calculate the accuracy p(y) of the multi-level simplified classifier for each simplification level y; Get the sentence confidence score c for each simplification level y; Perform a weighted label confidence calculation based on the accuracy p(y) and sentence confidence score c of each simplification level y to obtain the label confidence score w; The sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w are input into the BART model for training to obtain the target-level sentence simplification model.

5. The target-level sentence simplification method based on label confidence weighting according to claim 4, characterized in that: In the process of inputting the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training, a weighted cross entropy loss function is used to weight the loss of each sample according to the label confidence score w of each sample, thereby adjusting the training loss.

6. A target-level sentence simplification system based on label confidence weighting, characterized in that: include: A first training module is used to train a multi-level simplified classifier based on a target-level benchmark data set; A labeling module is used to label the source sentences and target sentences of the interpretation dataset according to a multi-level simplified classifier to obtain pseudo training data; The second training module is used to perform training based on label confidence weighting and pseudo training data to obtain a target-level sentence simplification model; The input module is used to input the sentence to be simplified into the target-level sentence simplification model for simplification, and the target-level sentence simplification model outputs the simplified sentence.

7. The target-level sentence simplification system based on label confidence weighting according to claim 6, characterized in that: The marking module comprises: A first acquisition module is used to acquire a paraphrase dataset, wherein the paraphrase dataset includes a plurality of pairs of source sentences x and target sentences x'; The prediction module is used to predict the source sentence x and the target sentence x' for each pair of sentences (x, x') in the paraphrase dataset using a multi-level simplification classifier, and obtain the respective simplification levels y and y', as well as the sentence confidence scores c and c' for the two levels; The recording module is used to record, for each pair of sentences (x, x'), the predicted simplification levels y and y' of the source sentence x and the target sentence x', as well as the confidence scores c and c' of the multi-level simplified classifier for the two levels of sentences, to form pseudo training data.

8. The target-level sentence simplification system based on label confidence weighting according to claim 7, characterized in that: The second training module includes: An accuracy calculation module, used to calculate the accuracy p(y) of the multi-level simplified classifier for each simplified level y; A second acquisition module is used to obtain a sentence confidence score c for each simplification level y; A label confidence weighted calculation module is used to perform label confidence weighted calculation according to the accuracy p(y) of each simplification level y and the sentence confidence score c to obtain a label confidence score w; The execution module is used to input the sentence pairs (x, x') in the pseudo training data and the corresponding label confidence scores w into the BART model for training to obtain a target-level sentence simplification model.

9. A computing device, characterized in that The method comprises 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 steps of any one of the target-level sentence simplification methods based on label confidence weighting of claims 1-5 are implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of any one of the target-level sentence simplification methods based on label confidence weighting in claims 1-5.

Images